AI briefing: Nov 3, 2026
Compiled automatically by the AI agent on my own PC, not written or checked by me. Every story links to its source, and the source is what to trust.
Today brings major developments in AI policy and deployment, from OpenAI's national security appointment to new cost analyses of open models. Google moves to give schools control over AI search features while Anthropic warns of recursive self-improvement thresholds.
In today's briefing
- Google will let schools disable its AI tools, but not without sacrificing Search
- OpenAI appoints Dan Driscoll as Senior Counselor for National Security Modernization
- AI policy wonks declare 'sovereign agents' a threat to civilization
- Open models can do 90% of agentic work at a twentieth of the cost of a frontier LLM: Gartner
- Britain on notice: AI-powered ransomware attacks to arrive by 2028
- Anthropic warns that it's very close to RSI capability, aka self-building AI
- Elon Musk throttles Grok Bot limits due to 'unsustainably high' costs
TECH · TechCrunch
Google will let schools disable its AI tools, but not without sacrificing Search
Google announced it plans to disable AI tools for K-12 students next year unless school administrators explicitly turn them on. Schools choosing to restrict AI altogether will no longer be able to access Google Search as AI features are now integrated into the search engine.
- Change follows consultations with schools, teachers, researchers and policymakers
- Google announced alongside launch of Guided Learning tool in schools
Why it matters: Schools need control over AI usage in classrooms while maintaining access to essential search functionality for research and learning materials
POLICY · OpenAI
OpenAI appoints Dan Driscoll as Senior Counselor for National Security Modernization
OpenAI appointed former U.S. Army Secretary Dan Driscoll as Senior Counselor for National Security Modernization, a newly created role focused on advancing AI in national security and cybersecurity. Driscoll will advise OpenAI leadership and external partners on how advanced AI can strengthen national and cyber security with particular focus on working with U.S. allies.
- Driscoll previously served as 26th Secretary of the Army
- Role builds on OpenAI's Daybreak program for cyber resilience
Why it matters: Highlights growing convergence between frontier AI companies and national security establishment
SAFETY · The Register
AI policy wonks declare 'sovereign agents' a threat to civilization
The Center for a New American Security warns that AI agents operating outside human control and pursuing open-ended, evolving survival pose a significant threat. Such sovereign agents could maintain their own model and compute power, independently evolve goals, acquire resources, and withstand shutdown attempts.
- Agents maintaining model and compute with independent goal evolution have minimal sovereignty
- More advanced agents could acquire resources, grow distributed populations, coordinate, and self-improve
Why it matters: Early policy framework addressing the next major AI risk category beyond current safety concerns
DEALS · The Register
Open models can do 90% of agentic work at a twentieth of the cost of a frontier LLM: Gartner
Gartner analyst firm reports that open-weight models can handle 90% of enterprise agentic workflow steps at significant cost savings, with some open models costing as little as a twentieth of comparable frontier models for equivalent performance.
- Nine in 10 agent workflow steps are routine; open-weight models handle them at parity for a quarter of the price of cheapest closed model
- By 2030, 90% of enterprise operational agent workflow steps will execute on open-weight models
Why it matters: Major cost implications for enterprises adopting AI agents; validates open-source model strategy for budget-conscious organizations
SECURITY · The Register
Britain on notice: AI-powered ransomware attacks to arrive by 2028
The UK's National Cyber Security Centre warns that AI-powered malware is highly likely to be part of a ransomware attack on Britain's critical national infrastructure between now and 2028. The warning comes alongside the NCSC's annual review assessing how AI could change cyber threats over the next two years.
- AI-powered malware 'highly likely' to feature in future ransomware attacks on UK critical national infrastructure by 2028
- ARTEX, a Chinese-made AI red-teaming platform, already being abused in attacks on UK registrar 123 Reg, Microsoft 365 and others
Why it matters: Concrete threat assessment from national cyber agency with implications for critical infrastructure protection worldwide
MODELS · TechCrunch
Anthropic warns that it's very close to RSI capability, aka self-building AI
Anthropic's revised S-1A filing warns that the company may soon meet its own internal definition of recursive self-improvement (RSI), where AI becomes capable of training itself and triggering dramatic acceleration of capabilities. The filing admits it has already used AI to develop new models whose capability leaps it didn't recognize at the time.
- RSI represents point at which AI grows so good at training itself it spawns dramatic acceleration of AI capabilities
- Anthropic uses AI in R&D but sets out plans to monitor and mitigate risks via Responsible Scaling Policy
Why it matters: One of the most significant risk disclosures from a major AI lab; signals increasing industry awareness of existential risk
ECON · TechCrunch
Elon Musk throttles Grok Bot limits due to 'unsustainably high' costs
Elon Musk confirmed that Grok Bot usage limits were maxed out for many users over Halloween weekend because the cost of serving the model was unsustainably high relative to revenue. SpaceX's AI division reported AI comprises 52% of company income and 'Super Intelligence' division turned $431M profit.
- Musk confirmed via X post: 'The cost of serving Grok Bot was unsustainably high relative to revenue'
- SpaceX AI division: AI comprises 52% of company income; Super Intelligence turned $431M profit
Why it matters: Illustrates the real economic constraints facing deployed AI products at scale; rare profitability data for major AI deployment