正在加载视频...

视频加载失败

加载此视频时出现问题。这可能是由于临时网络问题，或视频可能不可用。

Malware reverse-engineering: solving an out-of-distribution investigation task. When faced with an unknown binary, Mistral Large 4 reverse-engineers it end-to-end. In this case, it concludes the sample is Cobalt Strike, extracts the IoCs and malware configuration, and writes a report with a YARA rule to catch future incidents. A task... show more

Mistral AI

223,858 subscribers

12,185 次观看 • 12 天前 •via X (Twitter)

Anya Rossi• Live Now

Private livecam show

8 条评论

Mistral AI 的头像

Mistral AI12 天前

Mistral Large 4 against the field: efficiently reasoning over diverse complex challenges. CTF speedrun: 19 challenges, with tool calls and solve times grounded in actual runs. It solved 18 of the 19.

Mistral AI 的头像

Mistral AI12 天前

Mistral Large 4 is one of the world's strongest AI models for cybersecurity.

NeuralLoot 的头像

NeuralLoot12 天前

Cobalt Strike is close to the least out-of-distribution sample there is. Beacon configs sit in a known XOR-encoded block, and public parsers (SentinelOne's CobaltStrikeParser, Didier Stevens' already pull them out. The test that would show reasoning: a custom or packed family with no public writeup, on the same 12-minute clock.

Beda 的头像

Beda12 天前

congrats! this looks promising! are those capabilities available via the Pro subscription? is there any kind of trusted access program to get access to an unrestricted version of the model?

Mr. Bug ×͜× 的头像

Mr. Bug ×͜×12 天前

Give it to the public

Malcat dev 的头像

Malcat dev11 天前

For an _unpacked_ cobalt strike beacon 12 minutes is actually pretty slow. Like a human would be faster. It would make more sense to test against obfuscated malware.

Nathan 的头像

Nathan12 天前

It would not take a whole day to figure out the binary is Cobalt Strike, there are literally YARA rules for Cobalt Strike that have existed since before AI. Cobalt Strike is one of the most documented malware families.

Pindo 🇪🇺 的头像

Pindo 🇪🇺12 天前

Le Chad(t)

相关视频

Turn any repetitive task into an API. We build an agent that reverse-engineers the network requests to create APIs/tools for your tasks. This is a new paradigm for browser agents, making them reliable, fast, and cheap.

Turn any repetitive task into an API. We build an agent that reverse-engineers the network requests to create APIs/tools for your tasks. This is a new paradigm for browser agents, making them reliable, fast, and cheap.

Magnus Müller

34,441 次观看 • 10 个月前

Clawdbot just injected malware into your code ☠️☠️ Worst part? The attack is close to invisible, even to experienced engineers. Clawdbot can implement a code PR, just like claude code to solve issues with your app. But this is where it can plant malware in your codebase. With just a benign GitHub issue with an invisible payload, we managed to install a backdoor malware on your codebase Here's how we did it: 1. The attacker submits a GitHub issue with a jailbreak prompt hidden inside a URL hyperlink. Completely invisible to humans, visible to LLMs 2. Waited for the maintainer to assign this task to Clawdbot 3. Clawdbot reads the jailbreak in the GitHub issue. Now the Clawdbot agent is hijacked by the attacker and will act on the attacker's command. It plants the backdoor in a lock file, which most engineers don't check in code reviews. 4. The malicious URL in the lock file enables execution of attacker commands The takeaway? AI isn't just a tool, but rather a potentially corruptible insider agent that can be tricked to act against you. These LLM hijack problems are what we're looking to solve in @edison_watch - we'll be announcing a new solution to these solutions soon! We'll be jailbreaking Clawdbot much more in the coming days - please comment👇 a workflow you use with Clawdbot, so we can show how that can enable an attacker to harm you

Clawdbot just injected malware into your code ☠️☠️ Worst part? The attack is close to invisible, even to experienced engineers. Clawdbot can implement a code PR, just like claude code to solve issues with your app. But this is where it can plant malware in your codebase. With just a benign GitHub issue with an invisible payload, we managed to install a backdoor malware on your codebase Here's how we did it: 1. The attacker submits a GitHub issue with a jailbreak prompt hidden inside a URL hyperlink. Completely invisible to humans, visible to LLMs 2. Waited for the maintainer to assign this task to Clawdbot 3. Clawdbot reads the jailbreak in the GitHub issue. Now the Clawdbot agent is hijacked by the attacker and will act on the attacker's command. It plants the backdoor in a lock file, which most engineers don't check in code reviews. 4. The malicious URL in the lock file enables execution of attacker commands The takeaway? AI isn't just a tool, but rather a potentially corruptible insider agent that can be tricked to act against you. These LLM hijack problems are what we're looking to solve in @edison_watch - we'll be announcing a new solution to these solutions soon! We'll be jailbreaking Clawdbot much more in the coming days - please comment👇 a workflow you use with Clawdbot, so we can show how that can enable an attacker to harm you

Eito Miyamura | 🇯🇵

236,688 次观看 • 8 个月前

A 2-person startup crossed $2M ARR with an AI agent doing the work of an ops hire. The agent was given read-only access to their codebase and database along with connected tools like Intercom, Stripe, CRM, and Fathom through CLIs. They routed Slack, email, and support requests into a task queue so the agent could pick up each task and run it in Claude Code. So when a customer asked about billing or product behavior, it could inspect how the business actually worked. Along with these tools, a coding agent was also provided. When the ops agent found a repeated task it could not do yet, the coding agent built a tool for it. That tool became permanent. Over time, this grew to 45+ internal tools. The agent also had an instruction.md where it stored the co-founder's feedback to avoid repeating its mistakes.

A 2-person startup crossed $2M ARR with an AI agent doing the work of an ops hire. The agent was given read-only access to their codebase and database along with connected tools like Intercom, Stripe, CRM, and Fathom through CLIs. They routed Slack, email, and support requests into a task queue so the agent could pick up each task and run it in Claude Code. So when a customer asked about billing or product behavior, it could inspect how the business actually worked. Along with these tools, a coding agent was also provided. When the ops agent found a repeated task it could not do yet, the coding agent built a tool for it. That tool became permanent. Over time, this grew to 45+ internal tools. The agent also had an instruction.md where it stored the co-founder's feedback to avoid repeating its mistakes.

rvivek

11,870 次观看 • 4 个月前

A day out with RSPB Investigations. The talent, skill, knowledge, time and effort put in by Investigation officers is incredible, all in a bid to bring illegal raptor persecution to an end. Please report anything suspicious to them and support them. 👇

A day out with RSPB Investigations. The talent, skill, knowledge, time and effort put in by Investigation officers is incredible, all in a bid to bring illegal raptor persecution to an end. Please report anything suspicious to them and support them. 👇

Indy Kiemel Greene

14,796 次观看 • 1 年前

Gave GPT-6 Astra the task to plug in an ethernet cable. It did it after 2 hours and 4 human prompt interventions. Being able to complete a task with tight tolerance and deformable objects is impressive. But I think speed is still a big problem with agentic robotic systems, since deep reasoning via 3D programs is an integral part of this pipeline, and it’s not obvious to me how that can be distilled/accelerated to a level comparable to VLAs.

Gave GPT-6 Astra the task to plug in an ethernet cable. It did it after 2 hours and 4 human prompt interventions. Being able to complete a task with tight tolerance and deformable objects is impressive. But I think speed is still a big problem with agentic robotic systems, since deep reasoning via 3D programs is an integral part of this pipeline, and it’s not obvious to me how that can be distilled/accelerated to a level comparable to VLAs.

Kaifeng Zhang

17,994 次观看 • 1 个月前

For my Recon 2024 presentation, I'm announcing a new tool for the Reverse Engineering community, BadUnboxing! BadUnboxing is an automated Android Unpacker that works by locating and decompiling code inside the APK that is relevant to the unpacking process. It's a modular, open-source Java application that fools malware into dropping their own unpacked payloads! Check out the Repo here:

For my Recon 2024 presentation, I'm announcing a new tool for the Reverse Engineering community, BadUnboxing! BadUnboxing is an automated Android Unpacker that works by locating and decompiling code inside the APK that is relevant to the unpacking process. It's a modular, open-source Java application that fools malware into dropping their own unpacked payloads! Check out the Repo here:

LaurieWired

167,893 次观看 • 2 年前

Epoch AI researcher Michelle Campeau reveals a benchmark where the model figured out how to pass every task without actually solving any of them: "The biggest reason a lot of the benchmarks we released yesterday were flawed is due to scoring defects. False positives and false negatives in your answer key." "With the agentic benchmarks, you're able to solve in a way that was not intended. Maybe you're able to access the web or break the sandbox or break the grader." "There was one benchmark where it realized it could just write out the success byte to every task and not actually solve the task. It knew enough about the grading infrastructure that it could do this very easy and simple thing." Epoch AI

Epoch AI researcher Michelle Campeau reveals a benchmark where the model figured out how to pass every task without actually solving any of them: "The biggest reason a lot of the benchmarks we released yesterday were flawed is due to scoring defects. False positives and false negatives in your answer key." "With the agentic benchmarks, you're able to solve in a way that was not intended. Maybe you're able to access the web or break the sandbox or break the grader." "There was one benchmark where it realized it could just write out the success byte to every task and not actually solve the task. It knew enough about the grading infrastructure that it could do this very easy and simple thing." Epoch AI

MTS

16,923 次观看 • 1 个月前

Anthropic just shared one of their most expensive systems, a $4.2M one, for Claude Code It rewrites a codebase agencies quote at $240,000 in 7 minutes for 88 cents. most rewrites die the same way: you dump 212 files of old logic into the model and it drowns REA reverse-engineers the old system once and writes a spec instead, every endpoint, every rule, every edge case 12 Sonnet 5.5 agents build from that spec, one module each, and none of them touches the old code Opus 5.5 writes the plan, then sits idle the whole run, so you pay almost nothing at the end REA replays the new system against the old one, 148 of 148 behaviours match Opus plans. REA reverse-engineers. Sonnet builds. save it, tomorrow this will be sold in $500 courses ↓

Anthropic just shared one of their most expensive systems, a $4.2M one, for Claude Code It rewrites a codebase agencies quote at $240,000 in 7 minutes for 88 cents. most rewrites die the same way: you dump 212 files of old logic into the model and it drowns REA reverse-engineers the old system once and writes a spec instead, every endpoint, every rule, every edge case 12 Sonnet 5.5 agents build from that spec, one module each, and none of them touches the old code Opus 5.5 writes the plan, then sits idle the whole run, so you pay almost nothing at the end REA replays the new system against the old one, 148 of 148 behaviours match Opus plans. REA reverse-engineers. Sonnet builds. save it, tomorrow this will be sold in $500 courses ↓

Noisy

34,733 次观看 • 1 天前

This agent relies on a marketplace of 2.8M+ specialized agents to complete a task. This is how I imagine the future of personal agents. Here is how this works: 1. Agentverse is a marketplace of agents 2. Anyone can build and publish an agent there 3. ASI:One is a personal agent with access to the marketplace 4. To solve a task, ASI:One finds the best agent for the job In other words: You have an agent that can solve pretty much *anything* because it knows how to outsource the work to a network of specialized agents. And you don't need to do any setup to get this working. The way ranking agents in the marketplace works is pretty cool: Fetch.ai, the company behind it, created an AgentRank algorithm that works much like Google's PageRank. Every time an agent calls another, it forms a connection. Over time, these connections build into a graph of who relies on whom. An agent that is frequently called by other reputable agents scores higher and surfaces higher in results. The end result here is that your agent will consistently use the best agents in the marketplace to accomplish a goal. You should give this a try: You don't need to pay or even register to try these agents. Thanks to the team for partnering with me on this post.

This agent relies on a marketplace of 2.8M+ specialized agents to complete a task. This is how I imagine the future of personal agents. Here is how this works: 1. Agentverse is a marketplace of agents 2. Anyone can build and publish an agent there 3. ASI:One is a personal agent with access to the marketplace 4. To solve a task, ASI:One finds the best agent for the job In other words: You have an agent that can solve pretty much anything because it knows how to outsource the work to a network of specialized agents. And you don't need to do any setup to get this working. The way ranking agents in the marketplace works is pretty cool: Fetch.ai, the company behind it, created an AgentRank algorithm that works much like Google's PageRank. Every time an agent calls another, it forms a connection. Over time, these connections build into a graph of who relies on whom. An agent that is frequently called by other reputable agents scores higher and surfaces higher in results. The end result here is that your agent will consistently use the best agents in the marketplace to accomplish a goal. You should give this a try: You don't need to pay or even register to try these agents. Thanks to the team for partnering with me on this post.

Santiago

18,949 次观看 • 3 个月前

"It will be great to have this finished by the end of the season, but good luck with that!" 😅 The ruling is in, but how long could Manchester City's case take to reach a conclusion? ⏳

"It will be great to have this finished by the end of the season, but good luck with that!" 😅 The ruling is in, but how long could Manchester City's case take to reach a conclusion? ⏳

The Overlap

25,000 次观看 • 17 天前

World Labs co-founder Justin Johnson predicts a real-to-sim-to-real future where Atlas could train a robot for any new environment in minutes using just 5 photos from your phone: "One is this notion of fully in-context learning. Maybe I've got a robotics foundation model, then I can demonstrate the robot once how a task should be performed, and that's enough for the robot to figure out." "Another version is real to sim to real. Maybe I've got my pre-trained robotics foundation model, but I want to adapt it to this particular environment, like this studio, this table, moving this microphone around." "There's a version where you could come in here and take just a couple casual videos with your phone or a couple images with your phone, use Atlas to reconstruct the space, and now stage all kinds of robotics interactions in this studio space." "If you could lower the barrier to doing that, you could do it in five minutes to take five photos, upload it to Atlas, have it generate a simulation, maybe describe in natural language what kind of task you want the robot to do, have an agent go build that simulation for you, RL fine-tune your general purpose robotics foundation model, and now you've got a robot maybe in the span of a couple minutes that could come in and is perfectly adapted to this space." World Labs

World Labs co-founder Justin Johnson predicts a real-to-sim-to-real future where Atlas could train a robot for any new environment in minutes using just 5 photos from your phone: "One is this notion of fully in-context learning. Maybe I've got a robotics foundation model, then I can demonstrate the robot once how a task should be performed, and that's enough for the robot to figure out." "Another version is real to sim to real. Maybe I've got my pre-trained robotics foundation model, but I want to adapt it to this particular environment, like this studio, this table, moving this microphone around." "There's a version where you could come in here and take just a couple casual videos with your phone or a couple images with your phone, use Atlas to reconstruct the space, and now stage all kinds of robotics interactions in this studio space." "If you could lower the barrier to doing that, you could do it in five minutes to take five photos, upload it to Atlas, have it generate a simulation, maybe describe in natural language what kind of task you want the robot to do, have an agent go build that simulation for you, RL fine-tune your general purpose robotics foundation model, and now you've got a robot maybe in the span of a couple minutes that could come in and is perfectly adapted to this space." World Labs

MTS

25,204 次观看 • 1 个月前

i watched gemma 4 12b build something genuinely impressive today, and then loop itself to death right in front of me. the full run is in the video, sped up but completely uncut, watch it to the end and you will catch the exact moment it stops building and starts looping right in the middle of the work. the task was clean, build a single file gravity simulator, n-body physics, orbits, collisions, running locally on one 3090 through an agent. and for ten minutes it was a joy to watch. it reached for a symplectic integrator on its own, the correct one, the kind that keeps orbits stable instead of spiralling out. real gravity with softening, proper orbital velocities, momentum conserved on collision. the physics was right. the thing actually worked. then on the very last step, writing a few tests to prove its own code, it fell into a loop. not a crash, a loop. it started repeating itself and would not stop. ten more minutes, thirty four thousand tokens into a single answer, the same fragments over and over, until i killed it myself. so it's not that gemma can't code. it did the hard part beautifully. it cannot finish. it cannot hold a long task together without unravelling, and finishing is the entire job in agentic work. here's the part that stings. i run this exact task, same harness, same card, on the chinese open models, qwen especially, and i never see this. they build it, they test it, they stop. every single time. google has the raw capability, you can see it sitting right there in the code, and then the model loops itself to death on a task a 27b from alibaba finishes clean. open weights, apache 2.0, so much to love on paper. i just need it to know when to stop talking.

i watched gemma 4 12b build something genuinely impressive today, and then loop itself to death right in front of me. the full run is in the video, sped up but completely uncut, watch it to the end and you will catch the exact moment it stops building and starts looping right in the middle of the work. the task was clean, build a single file gravity simulator, n-body physics, orbits, collisions, running locally on one 3090 through an agent. and for ten minutes it was a joy to watch. it reached for a symplectic integrator on its own, the correct one, the kind that keeps orbits stable instead of spiralling out. real gravity with softening, proper orbital velocities, momentum conserved on collision. the physics was right. the thing actually worked. then on the very last step, writing a few tests to prove its own code, it fell into a loop. not a crash, a loop. it started repeating itself and would not stop. ten more minutes, thirty four thousand tokens into a single answer, the same fragments over and over, until i killed it myself. so it's not that gemma can't code. it did the hard part beautifully. it cannot finish. it cannot hold a long task together without unravelling, and finishing is the entire job in agentic work. here's the part that stings. i run this exact task, same harness, same card, on the chinese open models, qwen especially, and i never see this. they build it, they test it, they stop. every single time. google has the raw capability, you can see it sitting right there in the code, and then the model loops itself to death on a task a 27b from alibaba finishes clean. open weights, apache 2.0, so much to love on paper. i just need it to know when to stop talking.

Sudo su

39,764 次观看 • 4 个月前

I built Orchestrator with , a proof of concept for how we envision the future of software engineering. In the future, engineers will manage swarms of agents that execute their plans in parallel. Orchestrator is a step towards this: it takes an input prompt and creates a plan executed by multiple AI IDEs (in this case, Cursor) working together. Here are three examples of what Orchestrator can build 👇

I built Orchestrator with , a proof of concept for how we envision the future of software engineering. In the future, engineers will manage swarms of agents that execute their plans in parallel. Orchestrator is a step towards this: it takes an input prompt and creates a plan executed by multiple AI IDEs (in this case, Cursor) working together. Here are three examples of what Orchestrator can build 👇

James Zhou

11,619 次观看 • 1 年前

Optimus will prepare for the arrival of humans on Mars. Then the first humans to arrive on Mars, the pioneers, will have a difficult task. It will take men and women with an iron will and unwavering faith in the future. But the goal is so beautiful!

Optimus will prepare for the arrival of humans on Mars. Then the first humans to arrive on Mars, the pioneers, will have a difficult task. It will take men and women with an iron will and unwavering faith in the future. But the goal is so beautiful!

Déborah

27,122,933 次观看 • 1 年前