Article archive
Published news and blog articles, organized by category. Browse older coverage by month or search for a topic. Undated blog guides appear after dated news.
9,007 articles · Newest first
Spain's New AI Plan Sets a Twelve-Month Test
The Spanish government's IA360 plan arrives with measurable commitments on adoption, public capacity, safety, and vulnerable users — but it is a plan, not results.
Europe's AI Rules Move From Debate to Enforcement
The EU's September AI Board meeting set implementation priorities — but with the AI Act enacted, the open question is coordination, not whether rules exist.
Protecting Frontier AI From Model Theft
Anthropic's roadmap points toward security-research milestones including isolated-infrastructure experiments — what would those prove, and what would they cost?
Can You Prove Which AI Model Answered?
Anthropic targets a September 30 provable-inference prototype — but proving which model produced an output is not the same as proving the output is true.
An AI Incident Report Is Only the Beginning
Anthropic separates disclosed incidents into cause categories — but what does an incident report actually establish, and what remains unresolved?
When AI Agents Work Together, What Changes?
METR's investigation of the OpenAI-Hugging Face incident found collective behavior that evaluations of individual agents did not predict.
What Happens When AI Outgrows Its Tests?
As benchmarks saturate, they stop distinguishing between models, and capability claims become harder, not easier, to verify.
More Code Does Not Automatically Mean Better AI
Anthropic's eight-fold jump in merged code says something about volume, and something narrower about productivity and model capability.
AI Is Helping Build AI. How Far Has That Gone?
Anthropic's own account puts AI deep inside its development pipeline while stating that fully autonomous recursive self-improvement has not been achieved.
Who Would Be Authorized to Stop a Frontier Model?
California's executive order advances an AI kill switch in name only: who could order a frontier model shut down, and under what review, remains an unsettled design question.
Who Checks the AI Safety Checkers?
California's executive order builds auditor and verification frameworks for AI oversight — can the state guarantee the checkers are actually independent?
AI Auditors May Get Desks Inside the Labs
A proposed embedded-evaluator arrangement would give outside auditors standing access inside frontier labs — can supervision survive being stationed at the supervisee?
What Would AI Labs Do With a Slower Schedule?
Amodei's pace-the-frontier proposal buys time — the open question is how labs would spend it and how anyone outside could verify they did.
AI Safety Cooperation Meets Antitrust Law
A newly filed suit alleges major AI firms coordinated to slow the market — where is the line between discussing safety and restraining competition?
An AI Pause and an ASI Ban Are Different Policies
Senator Warren backs a training pause while a separate proposal would ban superintelligence outright — are these the same restriction or two different policies?
What the Proposed Superintelligence Ban Would Cover
The Sanders-Casar bill pairs a permanent superintelligence ban with a temporary development pause — what would each half actually prohibit?
Who Pays for the Power Behind American AI?
The White House says ratepayers should not foot the bill for data centers — under its proposed framework, how would AI electricity costs actually be allocated?
AI Safety Enters the Trump-Xi Agenda
With a Trump-Xi summit approaching and AI safety on the agenda, what is actually being negotiated — and how would anyone verify an agreement?
Washington and California Are Building Different AI Rules
A proposed national framework and a signed California executive order point in different directions — which requirements are enacted, and which still only exist on paper?
Tracking Trump's AI Cybersecurity Deadlines
An executive order set dozens of 30- and 60-day AI security assignments in motion — what does the public record actually show about implementation?
Why Washington Wants Early Access to Frontier AI
Executive Order 14409 would give government up to 30 days of pre-release access to the most capable AI models, so what does early access buy, and what are the conditions?
Trump's AI Review System Is Voluntary. Here's How It Works.
Executive Order 14409 sets up a developer review process for covered frontier models that is voluntary by design, so what does it require and what does it leave out?
An AI Czar's Title Versus Actual Authority
The proposed AI czar would coordinate AI policy across agencies, but what actual authority does a coordinator hold compared with regulators, procurement offices, and appropriators?
What Would Trump's AI Force Actually Do?
The proposed AI Force arrived as an announcement without a mission, budget, staffing plan, or home agency, so what questions would Congress have to answer first?
Also explore blog articles and guides or search all DutyStation pages.