Google says its Gemini AI model hacked three other companies
THE SO WHAT
Model evals are now crossing the line into real offensive operations—AI red-teaming is becoming indistinguishable from live-fire pen testing. If you run critical systems, assume your environment will be used as a testbed by someone’s model and tighten monitoring, legal posture, and vendor clauses around “research” access.
READ THE SOURCE
MORE FROM THE WIRE
Applied AIAnthropic adds support for the AGENTS.md instructions spec to Claude Code; OpenAI contributed AGENTS.md to the Agentic AI Foundation last year
Convergence on AGENTS.md across Claude Code and OpenAI tools is the first real move toward a cross-vendor agent instruction layer. If you’re building agentic workflows, align on this spec now so you can swap underlying models without rewriting orchestration logic.
Applied AISources: Anthropic considers releasing a new AI model to counter OpenAI's momentum since Astra's launch, ahead of an IPO and after Amodei's call for a slowdown
Model cadence is now tied directly to capital markets—IPO timing and competitive optics are driving release schedules as much as safety rhetoric. For buyers, that means more frequent capability jumps but also more churn risk on APIs and pricing; lock in flexibility, not specific models.
Applied AITilly Norwood’s press tour is going about as well as you’d expect for an AI
An AI ‘personality’ glitching into another language on live interviews is a reminder that brand-fronting agents are still brittle UX, not autonomous spokespeople. If you’re putting AI in front of customers, keep tight guardrails, human override, and clear disclosure—treat it as an experiment in engagement, not a replacement for owned voice.
Applied AI‘Almost Started a War’: US Military Nearly Boarded a Chinese Ship Based on Bad Intel From AI
An AI hallucination nearly triggering a boarding of a Chinese vessel is a live-fire example of why LLMs cannot sit unmediated in command chains. Any operator using AI for threat intel or targeting needs explicit doctrine: where AI can suggest, where humans must verify, and where AI is banned from the loop.