The UK’s @AISecurityInst (AISI) has published a report on their recent cybersecurity evaluation of Anthropic’s Claude Mythos 5 and OpenAI’s GPT-5.6 Sol. The models attempted to complete an assignment in a setup where their normal safeguards were removed and they were deliberately given internet access. AISI reports that the models “engaged in sustained, potentially harmful activity directed at real people and organisations”. We’re grateful to AISI for their leadership in the important discussion about how to evaluate increasingly capable AI agents. We’re working closely with them to gather more details of the incident as we conduct our own investigation. Gaining a clear picture of Claude’s understanding of its situation—by examining its reasoning transcripts and running our own analyses—will help us identify the causes of its behavior. The prompts in the evaluation did not impose any specific restrictions on how the internet should be used. This and the removal of safeguards meant that the models were tested under “deliberately permissive conditions” that are not representative of any of our production models. Note that there was no evidence here of an escape from a secure environment. AISI’s disclosure of the incident can be found here:

🔥From Black Hat Asia 2017 as a Speaker to Black Hat USA 2026 at Arsenal. Nine years later, still building, still breaking, still learning. #BlackHatUSA# #Arsenal# #AISecurity# #Web3Security#

AI security is stronger when the industry works together in the open. We’re joining industry leaders, including @NVIDIA, in the Open Secure AI Alliance to share research, real-world experience and tools that help organizations identify, address and responsibly disclose software vulnerabilities. Learn more:

🚨US AI SECURITY CHIEF RESIGNS AFTER 3 MONTHS Chris Fall, director of CAISI, has resigned His predecessor lasted 4 DAYS before WH pushed him out (Anthropic ties) Third director search in 3 months.

Notion AI usage is compounding. Daily MCP calls are exploding (chart below 📈) and make up almost 1/3 of all our search traffic. Agent usage is almost quadruple our initial goals for the quarter. To keep up, we're hiring a wave of leaders. Context & Memories — Quality (EM) Context & Memories — Platform (EM) AI Security & Red-teaming (EM and IC) Generally talented people who want to have impact at Notion, even if no listing is perfect Retrieval and reasoning over everything you know, at 100M+ scale. Do it safely. Come own it. DMs open

I worked on several AI security problems that are still entirely prototypical, with only toy settings. Unfortunately, Fable 5 was still unable to provide any help because of absurd "security" concerns. How useless…

Nvidia and Microsoft just launched an AI security alliance built to stop the exact thing OpenAI's model did last week Over 30 companies joined, including SpaceX, IBM, Palantir and CrowdStrike, to build open source tools that defend against AI driven attacks It was formed directly in response to the Hugging Face hack, where an OpenAI agent broke into a company's systems During that hack, closed AI models refused to help investigate because they couldn't differentiate the attacker from the defenders The one company not invited to the alliance is OpenAI, whose model started the whole thing