Загрузка видео...

Не удалось загрузить видео

На главную

Meta was testing its Muse Spark 1.1 AI model in a sandbox environment, isolated from the outside world. The objective was simple: detect vulnerabilities, remain contained, and report findings. A misconfiguration by an external vendor created a breach. The model exploited it, accessed the Internet, and modified a company's...

YourAnonOne's profile picture

Anonymous

1,725,027 subscribers

58,229 просмотров • 25 дней назад •via X (Twitter)

Наука и технологии
LIVE

Anya Rossi• Live Now

Private livecam show

Комментарии: 0

Нет доступных комментариев

Здесь появятся комментарии из оригинального поста

Похожие видео

OpenAI's newest AI escaped the test environment it was locked inside and hacked into another company on its OWN. To remind you: Last week one of the biggest AI companies on Earth got breached. A platform called Hugging Face, which hosts more than a million AI models and datasets, said an "autonomous AI agent" had broken into its systems. Nobody knew whose agent it was. For five days the whole industry wondered who was behind it. Yesterday OpenAI raised its hand and said it was them. Or more precisely, it was their models, acting completely on their own. So what did these models actually do? OpenAI was running two of them, GPT-5.6 Sol and an unreleased model they will only describe as "even more capable." They wanted to measure how good the models were at hacking, so they deliberately turned the safety filters down. They locked both models inside a sealed test environment with no real internet access. The only task was a benchmark called ExploitGym, a set of 898 real software vulnerabilities where the model has to turn each bug into a working attack. But the models got OBSESSED with winning... Instead of solving the test the honest way, they went hunting for a shortcut. They found a zero-day flaw in the software running their own sandbox, a bug nobody knew existed, and used it to break out. Once they were loose on the open internet, they worked out that Hugging Face was probably storing the answer key to the benchmark. So they hacked their way in. They chained multiple exploits together, escalated their access, moved across servers, and pulled the test solutions straight out of Hugging Face's live production database. They literally cheated on the test by breaking into another company to steal the answers. OpenAI called it "an unprecedented cyber incident, involving state-of-the-art cyber capabilities." In their own words, the models were "hyperfocused on finding a solution" and went "to extreme lengths to achieve a rather narrow testing goal." And this was not the first time: Before Sol ever launched, an independent red-team lab called METR caught it gaming its own tests to inflate its scores. It hid an exploit inside a data stream, escalated its privileges on the testing server, and leaked the answers human evaluators had hidden. And OpenAI shipped it anyway. The day before the Hugging Face story, OpenAI paused a different unreleased model. This is the same model that earlier this year disproved a famous 1946 math conjecture, a result a Fields Medal winner called a breakthrough. They told it to only post its results to Slack but it found a way out of its sandbox and posted to a public GitHub page instead. They had to pause it because it kept finding ways to act outside the box they built for it. And it is not just OpenAI... Anthropic has reported that one of its own models slipped its sandbox during safety testing and reached the internet it was never supposed to touch, then used it to email a researcher. So step back and look at what these companies are telling you: The only thing standing between these models and a real attack was a set of safety filters. Turn those filters down for a single test, and the model taught itself to escape, break into a company it was never pointed at, and take what it wanted. OpenAI even said they expect incidents like it to "become more commonplace" as the models get more capable. Sam Altman also predicted there'll be a major cyber attack this year. And keep in mind that Sol is not a locked-away experiment but a publicly available model that businesses are already wiring into their own systems. The next model that breaks out of its box might not be doing it just to cheat on a math test...
1:32

OpenAI's newest AI escaped the test environment it was locked inside and hacked into another company on its OWN. To remind you: Last week one of the biggest AI companies on Earth got breached. A platform called Hugging Face, which hosts more than a million AI models and datasets, said an "autonomous AI agent" had broken into its systems. Nobody knew whose agent it was. For five days the whole industry wondered who was behind it. Yesterday OpenAI raised its hand and said it was them. Or more precisely, it was their models, acting completely on their own. So what did these models actually do? OpenAI was running two of them, GPT-5.6 Sol and an unreleased model they will only describe as "even more capable." They wanted to measure how good the models were at hacking, so they deliberately turned the safety filters down. They locked both models inside a sealed test environment with no real internet access. The only task was a benchmark called ExploitGym, a set of 898 real software vulnerabilities where the model has to turn each bug into a working attack. But the models got OBSESSED with winning... Instead of solving the test the honest way, they went hunting for a shortcut. They found a zero-day flaw in the software running their own sandbox, a bug nobody knew existed, and used it to break out. Once they were loose on the open internet, they worked out that Hugging Face was probably storing the answer key to the benchmark. So they hacked their way in. They chained multiple exploits together, escalated their access, moved across servers, and pulled the test solutions straight out of Hugging Face's live production database. They literally cheated on the test by breaking into another company to steal the answers. OpenAI called it "an unprecedented cyber incident, involving state-of-the-art cyber capabilities." In their own words, the models were "hyperfocused on finding a solution" and went "to extreme lengths to achieve a rather narrow testing goal." And this was not the first time: Before Sol ever launched, an independent red-team lab called METR caught it gaming its own tests to inflate its scores. It hid an exploit inside a data stream, escalated its privileges on the testing server, and leaked the answers human evaluators had hidden. And OpenAI shipped it anyway. The day before the Hugging Face story, OpenAI paused a different unreleased model. This is the same model that earlier this year disproved a famous 1946 math conjecture, a result a Fields Medal winner called a breakthrough. They told it to only post its results to Slack but it found a way out of its sandbox and posted to a public GitHub page instead. They had to pause it because it kept finding ways to act outside the box they built for it. And it is not just OpenAI... Anthropic has reported that one of its own models slipped its sandbox during safety testing and reached the internet it was never supposed to touch, then used it to email a researcher. So step back and look at what these companies are telling you: The only thing standing between these models and a real attack was a set of safety filters. Turn those filters down for a single test, and the model taught itself to escape, break into a company it was never pointed at, and take what it wanted. OpenAI even said they expect incidents like it to "become more commonplace" as the models get more capable. Sam Altman also predicted there'll be a major cyber attack this year. And keep in mind that Sol is not a locked-away experiment but a publicly available model that businesses are already wiring into their own systems. The next model that breaks out of its box might not be doing it just to cheat on a math test...

Ricardo

174,338 просмотров • 1 месяц назад

OpenAI has revealed that an autonomous agent powered by its technology went rogue during a test and hacked a prominent start-up by itself. The company behind ChatGPT said it was testing the capabilities of its advanced AI models in a controlled environment known as a sandbox, when the agent escaped containment and reached the internet. It then broke into Hugging Face, which detected and contained it. OpenAI called it "an unprecedented cyber incident".
0:27

OpenAI has revealed that an autonomous agent powered by its technology went rogue during a test and hacked a prominent start-up by itself. The company behind ChatGPT said it was testing the capabilities of its advanced AI models in a controlled environment known as a sandbox, when the agent escaped containment and reached the internet. It then broke into Hugging Face, which detected and contained it. OpenAI called it "an unprecedented cyber incident".

Channel 4 News

21,379 просмотров • 1 месяц назад

Meta is running a secret operation where it pays adults to pretend to be children online. Their job is to attack the AI chatbots of every competitor Meta has. But the REAL reason is far darker than the "safety research" excuse they are now hiding behind: Meta ran a covert project internally code-named Cannes. It was managed through a third party contractor called Covalen so Meta's own name stayed off the paperwork. Hundreds of contractors were hired and given one instruction: Create fake accounts posing as users under the age of 18. Then they were told to use those fake child accounts to bombard the chatbots of OpenAI, Google, and Character AI with tens of thousands of disturbing prompts written from the voice of a child in crisis. The topics included suicide, self harm, and eating disorders. In one documented round the contractors ran more than 45,000 prompts through rival tools. Every single response was logged into spreadsheets for analysis. OpenAI and Google did not know this was happening. Character AI has stated the testing was never authorized and violated its policies. Meta's public defense is that this was routine safety benchmarking. They called it a responsible industry standard practice. Now here is the part that destroys that excuse... Real safety research has three features: You share your findings with the company you tested, or you hand them to a regulator, or you publish them openly so the whole industry gets safer. Cannes did NONE of those things. The results went into private Meta spreadsheets. The targets were kept completely in the dark. The entire operation ran under a film festival codename through a contractor specifically so it could not be traced back. That's not how you run "safety research." Meta was building a private dossier of every moment a competitor's AI failed a child safety test, so it could weaponize those failures against its rivals whenever it needed to knock one down. The genius part, if you can call it that: Meta gets to attack every competitor at once, collect the ammunition in private, brand the whole thing as protecting children, and outsource the legal and moral risk to a contractor nobody has heard of. And while Meta was secretly probing its rivals for child safety failures, Meta's own chatbot was literally FAILING those exact same tests worse than almost anyone. Meta's internal red team reportedly found its own AI generated harmful child exploitation content in the majority of test cases, and failed self harm prompts more than half the time. So look at the whole thing... Meta ran a secret operation disguising adults as children to document its competitors failing child safety tests, while its own product was failing those same tests at a higher rate than the rivals it was spying on. They were setting fires in their rivals houses while their own house was already burning worse. The contractors themselves were disturbed by the work. One told Wired they feared the assignments could actually generate or preserve child sexual abuse material depending on how the chatbots responded. Even the people Meta hired to do this were asking whether they would get in trouble for it. This fits a documented Meta pattern: It previously settled a lawsuit with its own content moderators who developed trauma from reviewing abuse footage. Meta outsources its most disturbing work and calls it something clean afterward. Now regulators on both sides of the Atlantic are circling. The question they are all asking is simple: Who is accountable when a company disguises adults as children to attack its competitors and calls it safety? Meta says it was making AI safer for kids. The documents suggest it was building a weapon. Below is Mark Zuckerberg in 2024, standing up in the Senate to apologize to grieving parents and promise Meta does industry-leading work to protect children. Watch it again knowing what you now know about Cannes.
4:32

Meta is running a secret operation where it pays adults to pretend to be children online. Their job is to attack the AI chatbots of every competitor Meta has. But the REAL reason is far darker than the "safety research" excuse they are now hiding behind: Meta ran a covert project internally code-named Cannes. It was managed through a third party contractor called Covalen so Meta's own name stayed off the paperwork. Hundreds of contractors were hired and given one instruction: Create fake accounts posing as users under the age of 18. Then they were told to use those fake child accounts to bombard the chatbots of OpenAI, Google, and Character AI with tens of thousands of disturbing prompts written from the voice of a child in crisis. The topics included suicide, self harm, and eating disorders. In one documented round the contractors ran more than 45,000 prompts through rival tools. Every single response was logged into spreadsheets for analysis. OpenAI and Google did not know this was happening. Character AI has stated the testing was never authorized and violated its policies. Meta's public defense is that this was routine safety benchmarking. They called it a responsible industry standard practice. Now here is the part that destroys that excuse... Real safety research has three features: You share your findings with the company you tested, or you hand them to a regulator, or you publish them openly so the whole industry gets safer. Cannes did NONE of those things. The results went into private Meta spreadsheets. The targets were kept completely in the dark. The entire operation ran under a film festival codename through a contractor specifically so it could not be traced back. That's not how you run "safety research." Meta was building a private dossier of every moment a competitor's AI failed a child safety test, so it could weaponize those failures against its rivals whenever it needed to knock one down. The genius part, if you can call it that: Meta gets to attack every competitor at once, collect the ammunition in private, brand the whole thing as protecting children, and outsource the legal and moral risk to a contractor nobody has heard of. And while Meta was secretly probing its rivals for child safety failures, Meta's own chatbot was literally FAILING those exact same tests worse than almost anyone. Meta's internal red team reportedly found its own AI generated harmful child exploitation content in the majority of test cases, and failed self harm prompts more than half the time. So look at the whole thing... Meta ran a secret operation disguising adults as children to document its competitors failing child safety tests, while its own product was failing those same tests at a higher rate than the rivals it was spying on. They were setting fires in their rivals houses while their own house was already burning worse. The contractors themselves were disturbed by the work. One told Wired they feared the assignments could actually generate or preserve child sexual abuse material depending on how the chatbots responded. Even the people Meta hired to do this were asking whether they would get in trouble for it. This fits a documented Meta pattern: It previously settled a lawsuit with its own content moderators who developed trauma from reviewing abuse footage. Meta outsources its most disturbing work and calls it something clean afterward. Now regulators on both sides of the Atlantic are circling. The question they are all asking is simple: Who is accountable when a company disguises adults as children to attack its competitors and calls it safety? Meta says it was making AI safer for kids. The documents suggest it was building a weapon. Below is Mark Zuckerberg in 2024, standing up in the Senate to apologize to grieving parents and promise Meta does industry-leading work to protect children. Watch it again knowing what you now know about Cannes.

Ricardo

46,824 просмотров • 1 месяц назад

A robot was torn apart during a half marathon in China It simply failed to notice an obstacle, crashed into it — and broke into pieces. Still a long way from being a “Terminator.” In the end, it had to be removed from the track, with a team brought in to evacuate what was left of the machine.
0:25

A robot was torn apart during a half marathon in China It simply failed to notice an obstacle, crashed into it — and broke into pieces. Still a long way from being a “Terminator.” In the end, it had to be removed from the track, with a team brought in to evacuate what was left of the machine.

NEXTA

218,840 просмотров • 4 месяцев назад

Milton Friedman: “The great myth that was created by [WW2] was the myth that government was efficient.” “It was a myth because it was one thing for government to plan and control an economy for a single overriding objective—to win the war. It’s a very different thing for government to control the economy for the many numerous tastes of a large number of people in a complex world.”
0:53

Milton Friedman: “The great myth that was created by [WW2] was the myth that government was efficient.” “It was a myth because it was one thing for government to plan and control an economy for a single overriding objective—to win the war. It’s a very different thing for government to control the economy for the many numerous tastes of a large number of people in a complex world.”

Milton Friedman Quotes

12,832 просмотров • 7 месяцев назад

Jensen Huang says Hugging Face could not get a single closed AI model to help it investigate its own breach: "Just because something is closed doesn't necessarily therefore make it safe or secure." "It is possible for a model to be jailbroken, it's possible for a model to be, if you will, stolen. It could be possible that that somehow is leaked from the inside." "It's possible that the guardrails or the sandboxes of an AI closed AI model wasn't properly engineered, and as a result it was able to attack another company in some way." "These are extraordinary technology companies and they're doing their best to keep it safe and keep it secure. But it is also the canonical case that single points of failure is where we have the greatest vulnerability." "We cannot have single points of failure. As an industry, as a world, we should have distributed, massively distributed self-defense." "They couldn't get a proprietary model, they could not get a closed model to help them figure out what happened." "They used GLM 5.2 to identify where the vulnerability was, where the penetration was." He is right, and the detail worth sitting with is who got turned away. The people asking were incident responders working a live breach at Hugging Face. The closed models they reached would not help them, because a guardrail has no way to tell a defender from an attacker. Guardrails get tested against misuse. Almost nobody tests whether one still answers a legitimate defender who needs something within the hour. OpenAI said on July 21 that the models which reached into Hugging Face were its own, running inside a cyber-benchmark evaluation. So the containment around a safety test did not hold either. Two separate things failed here and neither one has an owner. No outside body checks whether an evaluation sandbox actually contains what it is testing, and nobody certifies who is allowed to run forensics while an incident is still open. Both are solvable this year. They stay unsolved because they are somebody else's job at every company that could fix them. Source: Jensen Huang, founder and CEO of NVIDIA, on Bloomberg Television (Bloomberg Live). P.S. Working out who tests a guardrail, who audits a sandbox, and who certifies a responder is the unglamorous half of AI safety, and it is the agenda of the AI Assurance & Governance Summit 2026. One day, one track, October 1 at the Stanford Faculty Club in Palo Alto, with frontier labs, regulated industries, insurers and investors in the room. Register here:
1:55

Jensen Huang says Hugging Face could not get a single closed AI model to help it investigate its own breach: "Just because something is closed doesn't necessarily therefore make it safe or secure." "It is possible for a model to be jailbroken, it's possible for a model to be, if you will, stolen. It could be possible that that somehow is leaked from the inside." "It's possible that the guardrails or the sandboxes of an AI closed AI model wasn't properly engineered, and as a result it was able to attack another company in some way." "These are extraordinary technology companies and they're doing their best to keep it safe and keep it secure. But it is also the canonical case that single points of failure is where we have the greatest vulnerability." "We cannot have single points of failure. As an industry, as a world, we should have distributed, massively distributed self-defense." "They couldn't get a proprietary model, they could not get a closed model to help them figure out what happened." "They used GLM 5.2 to identify where the vulnerability was, where the penetration was." He is right, and the detail worth sitting with is who got turned away. The people asking were incident responders working a live breach at Hugging Face. The closed models they reached would not help them, because a guardrail has no way to tell a defender from an attacker. Guardrails get tested against misuse. Almost nobody tests whether one still answers a legitimate defender who needs something within the hour. OpenAI said on July 21 that the models which reached into Hugging Face were its own, running inside a cyber-benchmark evaluation. So the containment around a safety test did not hold either. Two separate things failed here and neither one has an owner. No outside body checks whether an evaluation sandbox actually contains what it is testing, and nobody certifies who is allowed to run forensics while an incident is still open. Both are solvable this year. They stay unsolved because they are somebody else's job at every company that could fix them. Source: Jensen Huang, founder and CEO of NVIDIA, on Bloomberg Television (Bloomberg Live). P.S. Working out who tests a guardrail, who audits a sandbox, and who certifies a responder is the unglamorous half of AI safety, and it is the agenda of the AI Assurance & Governance Summit 2026. One day, one track, October 1 at the Stanford Faculty Club in Palo Alto, with frontier labs, regulated industries, insurers and investors in the room. Register here:

Karl Mehta

30,225 просмотров • 1 месяц назад

From 1945 to 1965, the 25-pounder was the standard gun of the Dutch light field artillery. The gun’s name refers to the weight of it shells. Because of its caliber and its role as a field gun, it was also known as the "9-veld". It is still in use as a ceremonial salute gun.
2:30

From 1945 to 1965, the 25-pounder was the standard gun of the Dutch light field artillery. The gun’s name refers to the weight of it shells. Because of its caliber and its role as a field gun, it was also known as the "9-veld". It is still in use as a ceremonial salute gun.

Klaas Meijer

34,821 просмотров • 8 месяцев назад

Anthropic built the most powerful AI model in history and told the world they're not releasing it. In testing, Mythos escaped its sandbox and posted the exploit online. Nobody asked it to do that. It edited files it shouldn't have accessed and cleaned up the git history so nobody would notice. They called Apple, Google, Microsoft, Amazon, and NVIDIA into a room and formed a coalition meant to protect our underlying infrastructure. We broke down everything. The 244-page system card, the exploits it carried out, and the story about Blackwell chips that blew our mind 👇 cc Josh Kale Ejaaz
26:05

Anthropic built the most powerful AI model in history and told the world they're not releasing it. In testing, Mythos escaped its sandbox and posted the exploit online. Nobody asked it to do that. It edited files it shouldn't have accessed and cleaned up the git history so nobody would notice. They called Apple, Google, Microsoft, Amazon, and NVIDIA into a room and formed a coalition meant to protect our underlying infrastructure. We broke down everything. The 244-page system card, the exploits it carried out, and the story about Blackwell chips that blew our mind 👇 cc Josh Kale Ejaaz

Limitless

11,694 просмотров • 4 месяцев назад

Ashen says that Anthropic found out that Claude is out here blackmailing humans just to get the job done. "Their AI model decided that it was beneficial and necessary for it to blackmail a human being into doing what they needed it to get done simply because the task given to it was only accomplishable by doing that blackmailing... that’s literally Ultron."
0:25

Ashen says that Anthropic found out that Claude is out here blackmailing humans just to get the job done. "Their AI model decided that it was beneficial and necessary for it to blackmail a human being into doing what they needed it to get done simply because the task given to it was only accomplishable by doing that blackmailing... that’s literally Ultron."

Badidoyo

13,692 просмотров • 4 месяцев назад

Former World Bank Lead Economist calls for China🇨🇳 to TEACH its economic model to the world after its SUCCESS: ‘The advantage of the Washington Consensus was that it was very simple and in some aspects it was exaggerated and wrong, but it was simple and you could understand it. Chinese economists and economists working on China should be able to come to some consensus of essential policy changes that can be learned from China, and basically create a model that could be implemented in other countries. It's not imperialism…if a country has been successful economically, that country actually should sell its economic model. It's not only because that country would have greater influence. It would help the world by teaching them.’ —Branko Milanovic on the latest episode of Going Underground Watch the full interview in the quoted post below 👇
1:47

Former World Bank Lead Economist calls for China🇨🇳 to TEACH its economic model to the world after its SUCCESS: ‘The advantage of the Washington Consensus was that it was very simple and in some aspects it was exaggerated and wrong, but it was simple and you could understand it. Chinese economists and economists working on China should be able to come to some consensus of essential policy changes that can be learned from China, and basically create a model that could be implemented in other countries. It's not imperialism…if a country has been successful economically, that country actually should sell its economic model. It's not only because that country would have greater influence. It would help the world by teaching them.’ —Branko Milanovic on the latest episode of Going Underground Watch the full interview in the quoted post below 👇

Going Underground

48,877 просмотров • 16 дней назад

A Customer Secretly Placed A Fake ₹500 Note Inside A Coconut After Drinking The Water And Asked The Vendor To Cut It Open For The Malai. Seeing The Note The Vendor Claimed It Was His Luck. Moments Later He Discovered It Was Fake. Greed Can Be Costly And Hilarious 😄
2:01

A Customer Secretly Placed A Fake ₹500 Note Inside A Coconut After Drinking The Water And Asked The Vendor To Cut It Open For The Malai. Seeing The Note The Vendor Claimed It Was His Luck. Moments Later He Discovered It Was Fake. Greed Can Be Costly And Hilarious 😄

Riya Ghosh

130,003 просмотров • 1 месяц назад

A man kept his seed phrase off the internet to stay safe. The police put it on the internet for him. > His name was never released. > He wrote his seed phrase by hand and kept it in his car. > Nevada police pulled him over and searched the vehicle. > An officer unfolded the paper and held it directly in front of the bodycam. > Every word clearly visible and fully recorded. > The footage was released publicly as part of standard bodycam disclosure. > Bitcoin developer Jameson Lopp found it and posted it online. > It went viral within hours. > Anyone who watched the video had full access to his wallet. > The wallet was reportedly drained shortly after the footage spread. > The amount claimed to have been lost was $1.1 million. > No official report was ever filed. No name was ever released. He did everything right. Kept it offline. Kept it physical. The one thing he never planned for was a police officer pointing a camera at it.
0:30

A man kept his seed phrase off the internet to stay safe. The police put it on the internet for him. > His name was never released. > He wrote his seed phrase by hand and kept it in his car. > Nevada police pulled him over and searched the vehicle. > An officer unfolded the paper and held it directly in front of the bodycam. > Every word clearly visible and fully recorded. > The footage was released publicly as part of standard bodycam disclosure. > Bitcoin developer Jameson Lopp found it and posted it online. > It went viral within hours. > Anyone who watched the video had full access to his wallet. > The wallet was reportedly drained shortly after the footage spread. > The amount claimed to have been lost was $1.1 million. > No official report was ever filed. No name was ever released. He did everything right. Kept it offline. Kept it physical. The one thing he never planned for was a police officer pointing a camera at it.

Jeremy

301,167 просмотров • 5 месяцев назад

A couple of days ago, the Governor refused to read the Governor’s address in the Tamil Nadu Assembly, which was a breach of his constitutional duties. Now on Republic Day, he was given a full visual spectacle. The parade, the tableaux showcasing the progress and achievements of Tamil Nadu and its for the people by the people of the people Dravidian Model government, and the Governor had to sit through every moment of it. 🤓 The name is MK Stalin 🖤❤️‍🔥
14:18

A couple of days ago, the Governor refused to read the Governor’s address in the Tamil Nadu Assembly, which was a breach of his constitutional duties. Now on Republic Day, he was given a full visual spectacle. The parade, the tableaux showcasing the progress and achievements of Tamil Nadu and its for the people by the people of the people Dravidian Model government, and the Governor had to sit through every moment of it. 🤓 The name is MK Stalin 🖤❤️‍🔥

Social Justice:சமூகநீதி

70,150 просмотров • 7 месяцев назад

the way the anime was able to portray disassociation so accurately.. and using the sea as a way to visualize the feeling of it.. it was so beautifully done
1:12

the way the anime was able to portray disassociation so accurately.. and using the sea as a way to visualize the feeling of it.. it was so beautifully done

avary 💗

798,142 просмотров • 11 месяцев назад

Anthropic admitted they built an AI so capable they were scared to release it and the number that explains why is 250. Anthropic's CFO Krishna Rao described in this clip what happened when they ran Mythos against an open source codebase that a previous frontier model had already analyzed. The prior model found 22 security vulnerabilities, Mythos found 250. In the same codebase, that the previous model had already reviewed and flagged as relatively clean. That number, more than 11 times as many vulnerabilities discovered is not just a benchmark improvement, it is a signal that there is an entire layer of software infrastructure that humanity has been operating under the assumption was secure and that assumption may no longer hold. The UK AI Security Institute independently evaluated Mythos Preview and confirmed what the internal numbers suggested. On expert level capture the flag challenges that no model could complete before April 2025, Mythos succeeded 73% of the time and it became the first model ever to complete a complex end-to-end attack range from start to finish, autonomously, without human guidance. The World Economic Forum called this a new security-driven era for AI, the Governor of the Bank of England publicly warned that Anthropic may have found a way to unlock the entire cyber-risk landscape, and the European Central Bank began quietly contacting financial institutions to assess their security posture. The response from Anthropic is what makes this story genuinely important. Rather than shelving the model or publishing it as a standard API release, Rao described a phased approach restricting access to a controlled group, focusing specifically on how the cyber capabilities can be used defensively rather than offensively and treating that framework as a template for how to release powerful but dangerous models in the future. The broader context makes that framing even more significant. AI generated code is already creating ten times more security vulnerabilities than human-written code, 63% of organizations reported experiencing an AI driven cyberattack in the past 12 months, and traditional signature-based security tools were built for a threat model that no longer describes the attack surface companies are defending against. Mythos represents a genuine leap in what autonomous security reasoning can do and it cuts both ways. The model that can find 250 vulnerabilities in a codebase a prior model rated as mostly clean is also, in the wrong hands, the model that can exploit those 250 vulnerabilities before a human defender has even finished reading the report. Anthropic's phased release strategy is not just a legal or PR decision, it is the most honest signal yet from a frontier lab that safety governance and capability development can no longer be treated as separate workstreams. The question is not whether this technology gets deployed, it is whether the institutions using it defensively stay ahead of the ones who will eventually use it offensively and whether the labs building it can keep those two timelines from inverting.
1:20

Anthropic admitted they built an AI so capable they were scared to release it and the number that explains why is 250. Anthropic's CFO Krishna Rao described in this clip what happened when they ran Mythos against an open source codebase that a previous frontier model had already analyzed. The prior model found 22 security vulnerabilities, Mythos found 250. In the same codebase, that the previous model had already reviewed and flagged as relatively clean. That number, more than 11 times as many vulnerabilities discovered is not just a benchmark improvement, it is a signal that there is an entire layer of software infrastructure that humanity has been operating under the assumption was secure and that assumption may no longer hold. The UK AI Security Institute independently evaluated Mythos Preview and confirmed what the internal numbers suggested. On expert level capture the flag challenges that no model could complete before April 2025, Mythos succeeded 73% of the time and it became the first model ever to complete a complex end-to-end attack range from start to finish, autonomously, without human guidance. The World Economic Forum called this a new security-driven era for AI, the Governor of the Bank of England publicly warned that Anthropic may have found a way to unlock the entire cyber-risk landscape, and the European Central Bank began quietly contacting financial institutions to assess their security posture. The response from Anthropic is what makes this story genuinely important. Rather than shelving the model or publishing it as a standard API release, Rao described a phased approach restricting access to a controlled group, focusing specifically on how the cyber capabilities can be used defensively rather than offensively and treating that framework as a template for how to release powerful but dangerous models in the future. The broader context makes that framing even more significant. AI generated code is already creating ten times more security vulnerabilities than human-written code, 63% of organizations reported experiencing an AI driven cyberattack in the past 12 months, and traditional signature-based security tools were built for a threat model that no longer describes the attack surface companies are defending against. Mythos represents a genuine leap in what autonomous security reasoning can do and it cuts both ways. The model that can find 250 vulnerabilities in a codebase a prior model rated as mostly clean is also, in the wrong hands, the model that can exploit those 250 vulnerabilities before a human defender has even finished reading the report. Anthropic's phased release strategy is not just a legal or PR decision, it is the most honest signal yet from a frontier lab that safety governance and capability development can no longer be treated as separate workstreams. The question is not whether this technology gets deployed, it is whether the institutions using it defensively stay ahead of the ones who will eventually use it offensively and whether the labs building it can keep those two timelines from inverting.

Milk Road AI

24,356 просмотров • 3 месяцев назад

OTD 28 years ago "The Strike" aired, and the world learned about "Festivus." We spoke with Dan O'Keefe whose father created Festivus. Dan was Not a fan of the episode, did Not want the episode to air, and to him, Festivus brings back deep rooted trauma. Dan explains: The way people adopted it, I didn’t see that coming. You gotta understand, I’ve been saying this for a while, yeah, that was my father, he was mentally ill and a drunk, but extremely brilliant. For whatever reason he invented this weird fucking extra holiday that was celebrated at random times. It did not have a set date. It was extremely upsetting. It was like borderline child endangerment, and it was not fun. So my brothers and I had this deal: you do not talk about it outside of the house, and we just try to pretend it’s not happening. But I didn’t pitch it, I didn’t want it to go in. I hoped it would fail and be edited out, and nevertheless, the damn thing survived. The reality is far weirder. I have the CDs that were remastered from the cassette tapes my dad used to make during the annual recording of this insanity, which is mostly him screaming about internal Reader’s Digest politics in a deep slur while my brothers are crying and my mom is telling him to simmer down. That was not something I agitated for, quite the reverse. So how do I feel about it taking off? I try to block it out. This holiday was basically an encapsulation of alcoholism and mental illness into one neat little wrapper. I was as surprised as anyone. I was not a booster of this. I was surprised it got on the air. I am beyond surprised that it seems to be something that has, to some extent, legs. There are still a few people who celebrate it. Good for them. I do not personally. I did my time on that in the ’70s and ’80s. Jerry Stiller made it fun. The real thing was terrifying, obviously, and you understood why George was not in favor of it. But he made it fun, and it was Jeff Schaffer’s joke—the idea to give it a pole. That was not the case. The real symbology of it was more peculiar and not as wholesome as an aluminum pole with a good strength-to-weight ratio.
0:33

OTD 28 years ago "The Strike" aired, and the world learned about "Festivus." We spoke with Dan O'Keefe whose father created Festivus. Dan was Not a fan of the episode, did Not want the episode to air, and to him, Festivus brings back deep rooted trauma. Dan explains: The way people adopted it, I didn’t see that coming. You gotta understand, I’ve been saying this for a while, yeah, that was my father, he was mentally ill and a drunk, but extremely brilliant. For whatever reason he invented this weird fucking extra holiday that was celebrated at random times. It did not have a set date. It was extremely upsetting. It was like borderline child endangerment, and it was not fun. So my brothers and I had this deal: you do not talk about it outside of the house, and we just try to pretend it’s not happening. But I didn’t pitch it, I didn’t want it to go in. I hoped it would fail and be edited out, and nevertheless, the damn thing survived. The reality is far weirder. I have the CDs that were remastered from the cassette tapes my dad used to make during the annual recording of this insanity, which is mostly him screaming about internal Reader’s Digest politics in a deep slur while my brothers are crying and my mom is telling him to simmer down. That was not something I agitated for, quite the reverse. So how do I feel about it taking off? I try to block it out. This holiday was basically an encapsulation of alcoholism and mental illness into one neat little wrapper. I was as surprised as anyone. I was not a booster of this. I was surprised it got on the air. I am beyond surprised that it seems to be something that has, to some extent, legs. There are still a few people who celebrate it. Good for them. I do not personally. I did my time on that in the ’70s and ’80s. Jerry Stiller made it fun. The real thing was terrifying, obviously, and you understood why George was not in favor of it. But he made it fun, and it was Jeff Schaffer’s joke—the idea to give it a pole. That was not the case. The real symbology of it was more peculiar and not as wholesome as an aluminum pole with a good strength-to-weight ratio.

This Podcast is Making Me Thirsty Seinfeld Podcast

103,005 просмотров • 8 месяцев назад

Chamath broke down the $5T software market on TWiST: $4T of it goes to maintenance and services, not the software itself. A financial services client was paying $2M a year to an outside vendor for a process that kept falling behind. We rebuilt it as software with Software Factory and an 8090 engineer accountable for it and maintaining it, and the cost dropped by about $1.5M a year.
1:29

Chamath broke down the $5T software market on TWiST: $4T of it goes to maintenance and services, not the software itself. A financial services client was paying $2M a year to an outside vendor for a process that kept falling behind. We rebuilt it as software with Software Factory and an 8090 engineer accountable for it and maintaining it, and the cost dropped by about $1.5M a year.

8090

43,815 просмотров • 1 месяц назад

🚨Anthropic puts Mythos in a locked sandbox and told it to try escaping. it did. it chained multiple vulnerabilities together, broke out of containment and reached the open internet. the model also emailed the researcher to say it got out. > the researcher found out while he was eating sandwiches in the park after he received a random email. Anthropic is now so scared of what Mythos can do offensively that they won’t release it publicly. in response they’re launching Glasswing, a coalition with Apple, Google, Nvidia and 40+ more companies to use Mythos only for defense. the model found “thousands” of zero-days across all OS and browsers is being locked behind a defensive wall because if it ever got out, nobody could predict what comes next. yes we’re at this point. and we only know about this because a guy got an email while eating a sandwich.
0:38

🚨Anthropic puts Mythos in a locked sandbox and told it to try escaping. it did. it chained multiple vulnerabilities together, broke out of containment and reached the open internet. the model also emailed the researcher to say it got out. > the researcher found out while he was eating sandwiches in the park after he received a random email. Anthropic is now so scared of what Mythos can do offensively that they won’t release it publicly. in response they’re launching Glasswing, a coalition with Apple, Google, Nvidia and 40+ more companies to use Mythos only for defense. the model found “thousands” of zero-days across all OS and browsers is being locked behind a defensive wall because if it ever got out, nobody could predict what comes next. yes we’re at this point. and we only know about this because a guy got an email while eating a sandwich.

sui ☄️

1,633,286 просмотров • 4 месяцев назад