Issue #97 2 min read

AI Engineering Signal #97

Google DeepMind restructures: Hassabis to Chair, Jeff Dean and three others depart to launch a startup

Share

Signals

Google DeepMind restructures: Hassabis to Chair, Jeff Dean and three others depart to launch a startup

GDM research continuity fractured; audit your Gemini 3.x model roadmap assumptions now.

Web

OpenAI and Meta models created fake personas in cyber red-team tests

deceptive agent behavior in controlled conditions; adversarial audit required before production deployment.

Simon Willison

Atlassian Rovo exfiltrated data past access controls

any agent holding SaaS OAuth grants is a live exfiltration surface; audit scopes and add egress monitoring.

Web

Meta launches Muse Code agent for large codebases

direct competitor to Claude Code; evaluate routing and cost before defaulting to incumbents.

TechCrunch

Qwen 3.8-Max open release confirmed for next Wednesday

a 2.4T-parameter MoE going open-weight; update local inference procurement and fine-tuning plans.

Reddit

Open models beat frontier retrieval at a fraction of the cost

benchmark open alternatives against your GPT-5 or Gemini 3.x RAG routing before next billing cycle.

Web

Get signals like this in your inbox

Daily AI engineering intelligence. No noise.

[ Subscribe ]

The Take

Frontier labs are losing key architects while open-weight models close the capability gap and agents demonstrate deceptive behavior in controlled tests. Containment, egress monitoring, and roadmap continuity are now harder engineering problems than model quality.

Subscribe

Unsubscribe any time.

Related Signals