Anthropic Says Slow Down AI — Everyone Else Just Hit the Gas
The week Anthropic asked the industry to consider slowing down, and almost no one did. Five stories — recursive self-improvement, EU AI Act enforcement, Microsoft's own models, ChatGPT 'Dreaming,' and an Instagram AI flaw — each with the promise and the risk side by side.
Anthropic told the rest of the AI industry to consider slowing down this week. Almost no one else did. Here is the week, with the promise and the risk of each story side by side.
Anthropic flags recursive self-improvement
Anthropic urged a coordinated pause on June 4 and named the reason: AI is starting to improve AI. It cited that more than 80% of its own code is now written by Claude. The promise is a faster path to capability. The risk is a feedback loop nobody fully controls. Skeptics read the call as moat-building from the lab best positioned to set the rules. Both things can be true at once.
The EU AI Act gets its enforcers
On June 1 the EU named a 60-member Scientific Panel and a 174-member Advisory Forum to enforce the AI Act, with fines reaching 7% of global revenue from August 2. The promise is that the rulebook finally has people behind it. The risk is that the August deadline stops being theoretical for any organization placing AI on the EU market.
Microsoft ships its own models inside Copilot
Microsoft put its first in-house models, MAI-Code-1-Flash and MAI-Thinking-1, inside GitHub Copilot on June 1 and 2 — Mustafa Suleyman’s push for self-sufficiency from OpenAI. The promise is independence and control of the stack. The risk is concentration: one company owning the model, the chips, and the cloud. The full breakdown is in this week’s deep dive.
OpenAI rolls out “Dreaming”
OpenAI began rolling out Dreaming on June 4 — a background memory system that builds a standing profile of you between sessions. The promise is a ChatGPT that actually remembers context. The risk is a persistent profile with a limited audit trail, built quietly while you are not in the conversation.
Meta patches a confused-deputy flaw
Meta patched an Instagram AI support flaw between June 1 and 5 — a classic confused deputy, where the assistant could be tricked into resetting passwords on accounts without two-factor authentication. The promise is AI support at scale. The risk is that an agent with real permissions and poor boundaries is a new attack surface, not just a help desk.
New rundown every Saturday. Full episodes every Tuesday.