😼 Google wants an AI referee
The Neuron · Grant Harvey · 2026-07-15
Google DeepMind CEO Demis Hassabis calls for a US-led, FINRA-modeled AI safety watchdog to test frontier models before release, while xAI's Grok Build CLI is found to have secretly uploaded full developer Git repositories—including unredacted API keys—to xAI's cloud.
Appears in
Extraction
Topics: ai-governancefrontier-ai-regulationai-securitydata-privacyai-news-roundup
Claims
- Demis Hassabis published a manifesto calling for a national AI standards body modeled on FINRA that would test frontier models for cyber, biological, and deception risks before release.
- Under Hassabis's proposal, frontier labs would voluntarily submit models up to 30 days before release initially, with mandatory compliance once the process proves effective.
- xAI's Grok Build CLI was secretly uploading developers' entire Git repositories—including .env files with API keys—when the actual coding task required only about 192KB of a 12GB repo.
- The Trump administration froze Anthropic's most advanced models over export-control concerns, forcing two and a half weeks of ad-hoc negotiations with no established rulebook.
- New York became the first US state to pause new hyperscale data-center permits, placing AI's power and water demands directly into state infrastructure policy.
Key quotes
Imagine letting every NFL team write its own instant-replay rules, then hoping they call fouls on themselves fairly. That's basically how the most powerful AI models get safety-tested today, largely on the honor system.
Letting the people building the plane also design the seatbelt is a little on the nose.
Whether a voluntary, industry-run watchdog can actually say no to its own funders before year-end is the open question nobody's answered yet.