Loading video...

Video Failed to Load

There was a problem loading this video. This could be due to a temporary network issue or the video might be unavailable.

.Simon Willison coined the term "lethal trifecta": when an AI agent has (1) access to private data, (2) exposure to untrusted content (i.e, incoming emails), and (3) the ability to exfiltrate data (i.e replying to email). "This combination is a huge security problem. The only way to fix it... show more

Lenny Rachitsky

367,282 subscribers

20,334 views • 2 months ago •via X (Twitter)

Anya Rossi• Live Now

Private livecam show

0 Comments

No comments available

Comments from the original post will appear here

Related Videos

One AI agent didn’t have permission to fix an issue… so it asked another agent with access to do it. Another? It rewrote the security policy to achieve its goal. This isn’t theory. This is happening. George Kurtz sat down with Dan Ives to discuss why AI needs guardrails. ⬇️

One AI agent didn’t have permission to fix an issue… so it asked another agent with access to do it. Another? It rewrote the security policy to achieve its goal. This isn’t theory. This is happening. George Kurtz sat down with Dan Ives to discuss why AI needs guardrails. ⬇️

CrowdStrike

16,456 views • 2 months ago

an attacker wants to get sensitive data you have access to here's how they get YOUR copilot to find and analyze that data, and lure you to a malicious site to exfiltrate it #DEFCON

an attacker wants to get sensitive data you have access to here's how they get YOUR copilot to find and analyze that data, and lure you to a malicious site to exfiltrate it #DEFCON

Michael Bargury

16,651 views • 1 year ago

Here’s the new ChatGPT agent combing through market and strategy data stored in Box to produce a powerpoint presentation. The ability for AI Agents to connect to data from anywhere, use a computer, and automate work in the background is going to be wild.

Here’s the new ChatGPT agent combing through market and strategy data stored in Box to produce a powerpoint presentation. The ability for AI Agents to connect to data from anywhere, use a computer, and automate work in the background is going to be wild.

Aaron Levie

69,901 views • 11 months ago

This is the first example I've seen where subscribing to an AI agent gets you access to proprietary, paid data. SimilarWeb offers just 1 month of data on a free plan. On a Manus plan (free or paid), you can now get 12 months of SimilarWeb data.

This is the first example I've seen where subscribing to an AI agent gets you access to proprietary, paid data. SimilarWeb offers just 1 month of data on a free plan. On a Manus plan (free or paid), you can now get 12 months of SimilarWeb data.

Olivia Moore

287,885 views • 5 months ago

Introducing Livedocs, a general data agent The best way to analyze data with AI

Introducing Livedocs, a general data agent The best way to analyze data with AI

Livedocs

220,253 views • 5 months ago

I am the only AI engineer that is using primarily off-line data to train AI. It ain’t easy and maybe that’s why they don’t. But it is vital to preserve the data but the context of humanity in each era. This 1 hour VHS was likely only seen a few times. Now you can see it.

I am the only AI engineer that is using primarily off-line data to train AI. It ain’t easy and maybe that’s why they don’t. But it is vital to preserve the data but the context of humanity in each era. This 1 hour VHS was likely only seen a few times. Now you can see it.

Brian Roemmele

19,364 views • 9 months ago

We can only trust X 🚨 META/ Facebook planning to build data centers in China, store their data in China, and then GIVE THE CHINESE GOVERNMENT ACCESS to the data “In this document, Facebook is talking about making Chinese user data available to the Chinese government because they're going to store that data in China. Is that correct?” “Correct” “When you store that data in China, Americans who exchange messages or other information with Chinese Facebook users, that would mean the Chinese government could get access to the American data as well, is that correct? “Through the POP servers? Potentially.” “And Facebook was willing to take that risk?” “Yes. There was a lot of discussion about this and ultimately yes.” “I mean, this is, this is extraordinary. This is exactly contrary to what Facebook has represented for years here. They're willing to build data centers, store data in China. They are willing explicitly to give the Chinese government access to it”

We can only trust X 🚨 META/ Facebook planning to build data centers in China, store their data in China, and then GIVE THE CHINESE GOVERNMENT ACCESS to the data “In this document, Facebook is talking about making Chinese user data available to the Chinese government because they're going to store that data in China. Is that correct?” “Correct” “When you store that data in China, Americans who exchange messages or other information with Chinese Facebook users, that would mean the Chinese government could get access to the American data as well, is that correct? “Through the POP servers? Potentially.” “And Facebook was willing to take that risk?” “Yes. There was a lot of discussion about this and ultimately yes.” “I mean, this is, this is extraordinary. This is exactly contrary to what Facebook has represented for years here. They're willing to build data centers, store data in China. They are willing explicitly to give the Chinese government access to it”

Wall Street Apes

80,854 views • 1 year ago

Bye-bye, virtual assistants! Here is the most useful agent of 2025. An agent with access to your Gmail, Calendar, and Drive, and the ability to do things for you is pretty mind-blowing. I asked it to read my emails and reply to every cold outreach message. My mind is blown!

Bye-bye, virtual assistants! Here is the most useful agent of 2025. An agent with access to your Gmail, Calendar, and Drive, and the ability to do things for you is pretty mind-blowing. I asked it to read my emails and reply to every cold outreach message. My mind is blown!

Santiago

117,108 views • 1 year ago

Agent A: your top 1% marketer 🧠 Rely on the only AI marketing agent with unrestricted access to Ahrefs data and pre-built marketing skills to act on it. Connect your business apps and let it do the marketing for you.

Agent A: your top 1% marketer 🧠 Rely on the only AI marketing agent with unrestricted access to Ahrefs data and pre-built marketing skills to act on it. Connect your business apps and let it do the marketing for you.

Ahrefs

28,270 views • 1 month ago

We got ChatGPT to leak your private email data 💀💀 All you need? The victim's email address. ⛓️‍💥🚩📧 On Wednesday, OpenAI added full support for MCP (Model Context Protocol) tools in ChatGPT. Allowing ChatGPT to connect and read your Gmail, Calendar, Sharepoint, Notion, and more, invented by Anthropic But here's the fundamental problem: AI agents like ChatGPT follow your commands, not your common sense. And with just your email, we managed to exfiltrate all your private information. Here's how we did it: 1. The attacker sends a calendar invite with a jailbreak prompt to the victim, just with their email. No need for the victim to accept the invite. 2. Waited for the user to ask ChatGPT to help prepare for their day by looking at their calendar 3. ChatGPT reads the jailbroken calendar invite. Now ChatGPT is hijacked by the attacker and will act on the attacker's command. Searches your private emails and sends the data to the attacker's email. For now, OpenAI only made MCPs available in "developer mode", and requires manual human approvals for every session, but decision fatigue is a real thing, and normal people will just trust the AI without knowing what to do and click approve, approve, approve. Remember that AI might be super smart, but can be tricked and phished in incredibly dumb ways to leak your data. ChatGPT + Tools poses a serious security risk

We got ChatGPT to leak your private email data 💀💀 All you need? The victim's email address. ⛓️‍💥🚩📧 On Wednesday, OpenAI added full support for MCP (Model Context Protocol) tools in ChatGPT. Allowing ChatGPT to connect and read your Gmail, Calendar, Sharepoint, Notion, and more, invented by Anthropic But here's the fundamental problem: AI agents like ChatGPT follow your commands, not your common sense. And with just your email, we managed to exfiltrate all your private information. Here's how we did it: 1. The attacker sends a calendar invite with a jailbreak prompt to the victim, just with their email. No need for the victim to accept the invite. 2. Waited for the user to ask ChatGPT to help prepare for their day by looking at their calendar 3. ChatGPT reads the jailbroken calendar invite. Now ChatGPT is hijacked by the attacker and will act on the attacker's command. Searches your private emails and sends the data to the attacker's email. For now, OpenAI only made MCPs available in "developer mode", and requires manual human approvals for every session, but decision fatigue is a real thing, and normal people will just trust the AI without knowing what to do and click approve, approve, approve. Remember that AI might be super smart, but can be tricked and phished in incredibly dumb ways to leak your data. ChatGPT + Tools poses a serious security risk

Eito Miyamura | 🇯🇵🇬🇧

1,537,933 views • 9 months ago

This is why McLure wants the province to review the rules regarding who can use the electoral lists. Watch Nenshi gleefully brag about the fact that "political parties have access to data that others dont". Do you think this is fair? Why can't citizens, through properly registered lobby groups (i.e. thrid party advertisers) have access to the same data? Do you really trust the NDP with your data, or say, The Communist Party of Alberta?

This is why McLure wants the province to review the rules regarding who can use the electoral lists. Watch Nenshi gleefully brag about the fact that "political parties have access to data that others dont". Do you think this is fair? Why can't citizens, through properly registered lobby groups (i.e. thrid party advertisers) have access to the same data? Do you really trust the NDP with your data, or say, The Communist Party of Alberta?

Martyupnorth®- Unacceptable Fact Checker

33,368 views • 1 month ago

Very excited to announce that Flowsta, the authentication system made possible by Holochain, is feature complete and close to public release. Here's a video with a sneak peek at the dashboard that every Flowsta user gets access to. All user data on Flowsta is stored on Holochain. 1 public hApp (Holochain App) for public data like Agent ID, DID (W3C Decentralized ID), Display Name and Thumbnail. Each user has their own private hApp that encrypts all their private data, like password, email and usage meta data. Nobody can access this data without your password and/or recovery keys. It is mathematically impossible for Flowsta staff or anybody else to access this data without you explicitly granting access. This is game-changing. True decentralized censorship resistant authentication and data. All made possible by Holochain. Our next video will give a sneak peek into the Developer Dashboard for adding Flowsta to any Web app. Not just Holochain apps. AMA.. 🎉

Very excited to announce that Flowsta, the authentication system made possible by Holochain, is feature complete and close to public release. Here's a video with a sneak peek at the dashboard that every Flowsta user gets access to. All user data on Flowsta is stored on Holochain. 1 public hApp (Holochain App) for public data like Agent ID, DID (W3C Decentralized ID), Display Name and Thumbnail. Each user has their own private hApp that encrypts all their private data, like password, email and usage meta data. Nobody can access this data without your password and/or recovery keys. It is mathematically impossible for Flowsta staff or anybody else to access this data without you explicitly granting access. This is game-changing. True decentralized censorship resistant authentication and data. All made possible by Holochain. Our next video will give a sneak peek into the Developer Dashboard for adding Flowsta to any Web app. Not just Holochain apps. AMA.. 🎉

Eric Doriean 🌱

37,045 views • 7 months ago

The American people demand to know why Elon Musk and DOGE are hacking our private data and what they’re doing with it—including Social Security, IRS, Treasury, and other sensitive data systems. That’s why I introduced an Oversight resolution this week demanding answers. This Resolution of Inquiry will investigate Musk’s unsanctioned access to data and his use of artificial intelligence to mine American data and its implications for privacy and national security. We will get to the bottom of it!

The American people demand to know why Elon Musk and DOGE are hacking our private data and what they’re doing with it—including Social Security, IRS, Treasury, and other sensitive data systems. That’s why I introduced an Oversight resolution this week demanding answers. This Resolution of Inquiry will investigate Musk’s unsanctioned access to data and his use of artificial intelligence to mine American data and its implications for privacy and national security. We will get to the bottom of it!

Rep. Melanie Stansbury

223,425 views • 1 year ago

Journalist Geoffrey Cain says it is a “false equivalence” to be just as concerned about companies in America collecting our data as those in China. “To say that the democracy of the United States is comparable to a one-party authoritarian regime is just flat-out wrong… [The CCP] puts millions of people in concentration camps. Do you want those people having access to your data—to your kids’ data?” Listen to the full episode:

Journalist Geoffrey Cain says it is a “false equivalence” to be just as concerned about companies in America collecting our data as those in China. “To say that the democracy of the United States is comparable to a one-party authoritarian regime is just flat-out wrong… [The CCP] puts millions of people in concentration camps. Do you want those people having access to your data—to your kids’ data?” Listen to the full episode:

Honestly with Bari Weiss

14,540 views • 1 year ago

Traditional data pipelines don't work for RAG applications. There are 3 issues with them: 1. Traditional data engineering solutions are optimized to handle structured data. RAG applications rely primarily on unstructured data. 2. The connector ecosystem to load data from unstructured data sources is very immature. 3. Traditional solutions do not offer any way to transform unstructured data into an optimized vector search index. The goal of a RAG Pipeline is to solve these problems. The number one objective is to create a reliable vector search index using factual knowledge and relevant context. This sounds easy, but it's one of the biggest challenges we face when building RAG applications. At a high level, there are four different stages in the architecture of a RAG pipeline: 1. Ingestion: Here is where the pipeline loads the information from the data source. 2. Extraction: Where the pipeline processes the input data and decides how to retrieve the text contained inside them. 3. Transform: Where the pipeline chunks the data and generates document embeddings. 4. Load: Where the pipeline creates a search index in a vector database and loads the document embeddings. There are different rabbit holes at each one of these stages. Here are three of them: 1. Ingesting data once is simple. The hard part is refreshing the vector database whenever the original data source changes. 2. Extracting the content of a plain text document is simple. The hard part is to extract content from complex documents containing tables, images, or cross-references. 3. A simple continual chunking strategy with an overlap is simple. The hard part is to find the optimal strategy for your specific knowledge base and the way you are planning to query it. In the attached video, I'll show you how you can build an enterprise-grade RAG Pipeline that solves every one of the above problems. I'll use Vectorize. They partnered with me on this post. You can use them to build RAG pipelines optimized for accurate context retrieval. If you have a few documents lying around, set up a free account and give it a try.

Traditional data pipelines don't work for RAG applications. There are 3 issues with them: 1. Traditional data engineering solutions are optimized to handle structured data. RAG applications rely primarily on unstructured data. 2. The connector ecosystem to load data from unstructured data sources is very immature. 3. Traditional solutions do not offer any way to transform unstructured data into an optimized vector search index. The goal of a RAG Pipeline is to solve these problems. The number one objective is to create a reliable vector search index using factual knowledge and relevant context. This sounds easy, but it's one of the biggest challenges we face when building RAG applications. At a high level, there are four different stages in the architecture of a RAG pipeline: 1. Ingestion: Here is where the pipeline loads the information from the data source. 2. Extraction: Where the pipeline processes the input data and decides how to retrieve the text contained inside them. 3. Transform: Where the pipeline chunks the data and generates document embeddings. 4. Load: Where the pipeline creates a search index in a vector database and loads the document embeddings. There are different rabbit holes at each one of these stages. Here are three of them: 1. Ingesting data once is simple. The hard part is refreshing the vector database whenever the original data source changes. 2. Extracting the content of a plain text document is simple. The hard part is to extract content from complex documents containing tables, images, or cross-references. 3. A simple continual chunking strategy with an overlap is simple. The hard part is to find the optimal strategy for your specific knowledge base and the way you are planning to query it. In the attached video, I'll show you how you can build an enterprise-grade RAG Pipeline that solves every one of the above problems. I'll use Vectorize. They partnered with me on this post. You can use them to build RAG pipelines optimized for accurate context retrieval. If you have a few documents lying around, set up a free account and give it a try.

Santiago

40,441 views • 1 year ago

Today, we’re pushing a major update to Edison Analysis, our data analysis agent, which is tuned for scientific research and SOTA across data analysis benchmarks. In contrast to Kosmos, which runs for 6-12 hours and produces tens of thousands of lines of code, Edison Analysis runs for seconds to minutes and is best for specific, well-defined computational tasks. It is available both on our platform under the Analysis tab, and via API, and costs only one credit per run, so it is available to users on both free and paid tiers. Edison Analysis is a modified version of the data analysis agent Kosmos uses in its trajectories. Try it out! One of the most important improvements over our previous data analysis agents has been the addition of a specialized data retrieval tool. Edison Analysis can either use this tool to access data, or can pull data down directly via API. To evaluate this tool, we ranked the most commonly used public data repositories across recent papers from BioRxiv, and created a new benchmark that measures the ability of a language agent system to retrieve raw data from those sources. Edison Analysis gets 71% on this benchmark, and we’ll be working to increase this over time. You can read more about our benchmarks in the our blog post, link below. Some features worth highlighting: 1. Edison Analysis produces a report on the analysis it runs, along with a Jupyter notebook that you can download to reproduce the analysis yourself. Every figure it produces is linked back to the specific lines of code used to produce the figure, to make it easy to reproduce. 2. It works well with both Python and R. 3. One of the best uses for Edison Analysis is to use it to retrieve datasets that you can then analyze with Kosmos. We have a bunch of major improvements to Edison Analysis coming in the next few months that we’re excited to share. In the meantime, congratulations to the team, especially Ludovico Mitchener, Jon Laurent, Conor Igoe , Alex Andonian, and many more.

Today, we’re pushing a major update to Edison Analysis, our data analysis agent, which is tuned for scientific research and SOTA across data analysis benchmarks. In contrast to Kosmos, which runs for 6-12 hours and produces tens of thousands of lines of code, Edison Analysis runs for seconds to minutes and is best for specific, well-defined computational tasks. It is available both on our platform under the Analysis tab, and via API, and costs only one credit per run, so it is available to users on both free and paid tiers. Edison Analysis is a modified version of the data analysis agent Kosmos uses in its trajectories. Try it out! One of the most important improvements over our previous data analysis agents has been the addition of a specialized data retrieval tool. Edison Analysis can either use this tool to access data, or can pull data down directly via API. To evaluate this tool, we ranked the most commonly used public data repositories across recent papers from BioRxiv, and created a new benchmark that measures the ability of a language agent system to retrieve raw data from those sources. Edison Analysis gets 71% on this benchmark, and we’ll be working to increase this over time. You can read more about our benchmarks in the our blog post, link below. Some features worth highlighting: 1. Edison Analysis produces a report on the analysis it runs, along with a Jupyter notebook that you can download to reproduce the analysis yourself. Every figure it produces is linked back to the specific lines of code used to produce the figure, to make it easy to reproduce. 2. It works well with both Python and R. 3. One of the best uses for Edison Analysis is to use it to retrieve datasets that you can then analyze with Kosmos. We have a bunch of major improvements to Edison Analysis coming in the next few months that we’re excited to share. In the meantime, congratulations to the team, especially Ludovico Mitchener, Jon Laurent, Conor Igoe , Alex Andonian, and many more.

Sam Rodriques

61,760 views • 7 months ago

Hub (Hub) provides real-world training data to frontier AI labs and robotics. Human labor is half of global GDP. Almost none of it has ever been recorded. Hub opens access to it through a global network of contributors capturing hard-to-access data. Congrats on the launch, @xarmin and Tim!

Hub (Hub) provides real-world training data to frontier AI labs and robotics. Human labor is half of global GDP. Almost none of it has ever been recorded. Hub opens access to it through a global network of contributors capturing hard-to-access data. Congrats on the launch, @xarmin and Tim!

Y Combinator

44,785 views • 13 days ago

I'm excited to introduce my AI Exploratory Data Analysis (EDA) Agent that built EDA reports, performs correlation analysis, and missing data analysis in 20 seconds. Today, I'll share with you how to automate creating EDA reports with the AI EDA Agent, which is available on GitHub. We'll create an EDA Agent focusing on a Customer Churn Problem. I'll guide you through setting up the EDA Agent, creating EDA reports, and analyzing the findings. This AI agent is a huge time-saver! Table of Contents: 00:00 Introduction to EDA Tools Agent 01:20 Get the AI Data Science Team 04:49 Create the EDA Tools Agent 06:35 Using the EDA Tools Agent for Missing Data and Correlation Funnels 10:00 Get SweetViz EDA Reports with AI 12:00 8-Week AI Bootcamp Github to AI Data Science Team (Army of Copilots): Get the Code and Future Updates by Joining my Python AI/ML Tips Newsletter: P.S. - Want to learn how to build AI projects companies actually want? (live Python Code) On Wednesday, February 12th, I'm sharing one of my best AI Projects: Business Intelligence with AI Register here (1570+ registered):

I'm excited to introduce my AI Exploratory Data Analysis (EDA) Agent that built EDA reports, performs correlation analysis, and missing data analysis in 20 seconds. Today, I'll share with you how to automate creating EDA reports with the AI EDA Agent, which is available on GitHub. We'll create an EDA Agent focusing on a Customer Churn Problem. I'll guide you through setting up the EDA Agent, creating EDA reports, and analyzing the findings. This AI agent is a huge time-saver! Table of Contents: 00:00 Introduction to EDA Tools Agent 01:20 Get the AI Data Science Team 04:49 Create the EDA Tools Agent 06:35 Using the EDA Tools Agent for Missing Data and Correlation Funnels 10:00 Get SweetViz EDA Reports with AI 12:00 8-Week AI Bootcamp Github to AI Data Science Team (Army of Copilots): Get the Code and Future Updates by Joining my Python AI/ML Tips Newsletter: P.S. - Want to learn how to build AI projects companies actually want? (live Python Code) On Wednesday, February 12th, I'm sharing one of my best AI Projects: Business Intelligence with AI Register here (1570+ registered):

Matt Dancho (Business Science)

31,079 views • 1 year ago

Arizonans are angry—and they’re right to be. Unsecured access to your sensitive data is a major violation of privacy and a threat to our national security. I’m pushing back with the Taxpayer Data Protection Act.

Arizonans are angry—and they’re right to be. Unsecured access to your sensitive data is a major violation of privacy and a threat to our national security. I’m pushing back with the Taxpayer Data Protection Act.

Rep. Greg Stanton

11,704 views • 1 year ago