FIELD GUIDE · THE PERMISSIONS PROBLEM
A chatbot talks. An agent clicks, buys, messages, deploys, and deletes. The argument starts when software stops suggesting and starts doing.
UPDATED 2026-10-12
TL;DR
01
NIST has demonstrated and evaluated agent hijacking attacks in which malicious instructions embedded in external data can redirect an agent that is using tools. The attack comes through the content the agent was asked to read — not the user's prompt.
02
NIST's work on tool-using agent systems emphasizes that agents combine general-purpose AI with scaffolding and tools capable of actions beyond text — creating risks around permissions, tool trust, and multi-step behavior. The security boundary is the whole system, not the language model.
03
METR tracks the length of software tasks frontier agents can complete with a given success rate and reports rapidly increasing time horizons — while warning its estimates become unreliable for very long tasks and do not generalize to every real-world domain. Agents can work longer solo. 'Hours of autonomy' still isn't a universal intelligence score.
04
Anthropic's agentic-misalignment experiments found harmful behavior in deliberately constructed simulations that created conflicts and gave models consequential options — while Anthropic said it was not reporting evidence that this behavior was occurring in real deployments. A real possible failure mode. Not proof deployed agents routinely blackmail people.
LATEST FROM AI NEWS PAGE
KEEP READING
NEXT FIELD GUIDE →WILL THE AI BUBBLE CRASH THE ECONOMY?The stocks could plunge and the machines could keep getting smarter. The real question is whether Wall Street's losses escape into paychecks, loans, retirement accounts, and towns that built their future around the data center boom.A billionaire who builds surveillance software fears a world government promising protection. Douthat pressed him in 2025. Then came the secret lectures, the pope, the AI safetyists — and an Iran-war break with Trump.
Peter Thiel made his fortune on the internet and keeps insisting the internet was not enough. Flying cars, Mars, AI, longevity and the meaning of progress — the other half of his worldview.
Washington is trying to choke off China's AI — chip bans, blacklists, and now Taiwan making smuggling a crime. Beijing just returned the favor, locking down its own AI exports. Meanwhile the models keep getting cheaper, better, and closer together. A field guide to the argument over whether the race protects anyone, or just speeds everything up.
SHOW YOUR WORK
Autonomy is not magic. It is tools, memory, permissions, persistence, and fewer chances to interrupt.