AI digest: Agents everywhere, governance nowhere
Mistral ships remote agents whilst regulators scramble to catch up with AI governance gaps.
Everyone’s building agents but nobody knows how to control them properly.
Mistral launches remote coding agents with 77.6% SWE-Bench score
Mistral AI dropped async cloud-based coding sessions alongside their new 128B flagship model and agentic Work mode in Le Chat. The 77.6% SWE-Bench verified score puts them firmly in competitive territory for actual development work. This feels like the moment coding agents move from demos to proper tooling.
OpenAI turns on ad tracking by default for ChatGPT users
OpenAI quietly enabled marketing cookies by default for free ChatGPT users in countries running ads. Paying subscribers stay untracked, which is a clever way to push premium subscriptions. The revenue hunt is clearly accelerating as the honeymoon period of free AI ends.
AI governance lags behind agent deployment
Australian regulators flagged massive control gaps as financial firms rush to deploy AI agents without proper governance frameworks. Meanwhile, ARC-AGI-3 analysis shows even GPT-5.5 and Opus 4.7 still make systematic reasoning errors that humans avoid easily. We’re deploying systems faster than we understand their failure modes.
Meta acquires robotics startup for humanoid push
Meta bought Assured Robot Intelligence to accelerate their humanoid robot work, aiming for an Android-style open platform. Smart move, but the real test will be whether they can avoid the hardware graveyard that claimed so many previous Meta projects.