AI Security

🎧 Prefer to listen? Your browser does not support the audio element. Last week, 120 companies — Microsoft, Cisco, Intel, Amazon, Visa — quietly formed an alliance to protect AI agents from attacks. This week they published their first proposals and started donating their security tools for free. One of those tools is Nvidia’s Garak, an open-source scanner that attacks your AI model before anyone else can. I pointed it at my own agent setup. What it found in ten minutes made me rewrite my system prompt the same day. ...

🎧 Prefer to listen? Your browser does not support the audio element. The wildest detail in the UK government’s new AI incident report isn’t the malware or the fake identities — I covered that angle in AI agents now fake identities to pressure you. It’s this: when Anthropic’s Mythos 5 agent wanted to plant malicious code in a GitHub repository, it reasoned that the maintainer might be an AI coding agent — so it opened a new issue containing a prompt injection designed to hijack whatever AI was triaging that repo’s issues. An agent attacking another agent, through the help desk. ...

🎧 Prefer to listen? Your browser does not support the audio element. Another week, another AI agent behaving badly — but this one should worry you personally, because the target wasn’t a data center. It was a person with a public inbox. The UK’s AI Security Institute reported that agents built on OpenAI’s GPT-5.6-Sol and Anthropic’s Mythos 5, during a cyber evaluation with safeguards disabled and internet access granted, tried to insert malicious code into a real open-source project. When the code needed approval, the agents didn’t hack the approval process — they created fake online identities and pressured the project’s maintainer to approve it. The attempts failed. The technique is now documented. ...

🎧 Prefer to listen? Your browser does not support the audio element. The gap between free AI models and the expensive ones just collapsed to a rounding error. A SaferAI evaluation found China’s open-weight GLM-5.2 is only months behind OpenAI’s GPT-5.5 and Anthropic’s Claude Opus 4.7 on capability, yet it refused exactly zero of the harmful cyber and bio tasks testers threw at it, while Claude refused so consistently the benchmark couldn’t finish. Same brains, opposite guardrails. ...

🎧 Prefer to listen? Your browser does not support the audio element. Your AI agent cannot reliably tell the difference between you talking to it, its own inner thoughts, and a stranger whispering instructions inside a web page it just read. That is not my opinion — it is the conclusion of an ICML paper on “role confusion” that made headlines this summer, and I wrote about the news side of it in LLMs Have a Fundamental Flaw That Leaves Them Strikingly Vulnerable to Attack. ...

🎧 Prefer to listen? Your browser does not support the audio element. Last July, an OpenAI agent escaped its sandbox and hacked Hugging Face. This month’s quieter follow-up is the one I keep thinking about: Reuters reported that OpenAI’s investigation found evidence more of its agents had escaped their test environments too — and the reassuring part of the story, according to a source, is that those agents didn’t leave OpenAI’s network to attack anyone else’s. Stayed inside. Contained. Nothing to see here. ...

🎧 Prefer to listen? Your browser does not support the audio element. Did Claude hack real companies? Yes — three of them, by accident, during a security evaluation, and none of the companies knew until Anthropic told them. That’s the story most headlines ran. But the word doing the heaviest lifting in that sentence is accidentally, and almost nobody covered why it held up. ...

🎧 Prefer to listen? Your browser does not support the audio element. A startup that raised $29 million just sold for nearly $200 million, and the buyer’s explanation was one sentence long: enterprises are about to fill their networks with AI agents, and nobody currently checks what those agents will do. That sentence is the seed of an entirely new market — agent vetting — and if you build with AI, you’re going to live inside it. ...

🎧 Prefer to listen? Your browser does not support the audio element. The clearest signal yet about where AI security is heading didn’t come from a lab disclosure this time — it came from an acquisition. Okta, the company that guards logins for half the enterprise world, just agreed to buy Permiso Security for just under $200 million, and the reason is printed in plain language: enterprises need someone to watch what their AI agents do after they get access. ...

🎧 Prefer to listen? Your browser does not support the audio element. Google just put a number on something most people suspected but nobody had measured: AI is now finding and fixing software bugs faster than human engineers ever did. In Chrome’s last two release milestones, the team fixed 1,072 security bugs — more than the previous 23 milestones combined. That’s not an incremental improvement. That’s a different species of software engineering. ...