AI Industry Daily News

Trump Says He No Longer Views Anthropic as a National Security Threat

Trump told Axios's Marc Caputo that Anthropic had addressed the administration's concerns "very quickly and responsibly," saying of the company being a threat: "Well, not now, but a week ago, maybe." The pre-recorded interview referenced a meeting with Amodei at the G7 summit. Anthropic said in a statement that it appreciates the administration's continued partnership.

The comments follow a rapid escalation a week earlier. On June 12, the US Commerce Department issued an export-control directive citing national security authorities, ordering Anthropic to cut off all foreign access — inside and outside the US, including its own employees — to its most advanced models, Claude Fable 5 and Claude Mythos 5. Anthropic responded by abruptly disabling both models for all users worldwide while it made modifications; other Claude models were unaffected.

The dispute traces to February–March 2026, when the Pentagon sought unrestricted use of Claude and Anthropic insisted on limits for mass surveillance and autonomous weapons. The Pentagon labeled Anthropic a "supply chain risk," the first US company to receive the designation. The June trigger was reportedly a "limited, non-universal jailbreak" that could identify known minor vulnerabilities in certain codebases; Anthropic argued the issue was limited and not unique to its models, calling the response excessive.

Japan Drafts First Revision of AI Basic Plan, Signaling AI Act Review

The draft, which the government aims to approve by Cabinet in July, cites frontier models such as Anthropic's Claude Mythos as a basis for new mechanisms to collect performance and risk information from developers and for institutional changes geared toward emergency response. It also strengthens the AI Safety Institute (AISI) to evaluate high-performance models, gather vulnerability information, set guidelines, and lead international governance discussions.

The AI Basic Plan, first established in December 2025, is Japan's first systematic national AI strategy and anticipates annual reviews. Its underlying AI Act, enacted in 2025, is a promotion-oriented law without penalties, focusing on advice and guidance, prompting ongoing debate over adding penalties for bad actors.

Officials cite rapidly advancing AI capabilities as the reason for the urgency. Anthropic's Claude Mythos, disclosed in April 2026, reportedly showed strong performance in finding system vulnerabilities and generating attack code — scoring 93.9% on SWE-bench Verified, 97.6% on USAMO, and 83.1% on CyberGym — raising misuse concerns at the political and administrative level.

OpenAI Publishes Research on Training Broadly and Robustly Beneficial Models

The work targets the challenge of maintaining safe behavior as AI systems are deployed on long-horizon, high-stakes tasks where they may operate outside the situations they were explicitly trained for.

The research adds to a broader industry focus on reliability in production use, including recent attention to a usage-limit bug that affected roughly 3% of Claude Code Max/Pro users.

OpenAI o3 Deep Research Used for Rare Genetic Disease Diagnosis in NEJM AI Paper

The study demonstrates the use of AI in specialized clinical domains, supporting clinicians in identifying rare hereditary conditions.

The publication marks continued progress in validating AI systems for medical and scientific applications.