Anthropic restricts Claude Mythos to Project Glasswing

// 46d agoMODEL RELEASE

Anthropic restricts Claude Mythos to Project Glasswing

Anthropic has announced Claude Mythos Preview, a high-risk model restricted to a defensive coalition named Project Glasswing due to its unprecedented autonomous hacking capabilities. This gated release triggered an emergency meeting between the US Treasury, the Federal Reserve, and top bank CEOs to address systemic cyber risks to the global financial system.

// ANALYSIS

The launch of Mythos signifies a pivot from AI as a chatbot to AI as a systemic threat, necessitating a controlled, infrastructure-first defensive posture. The model achieves a record 93.9% SWE-bench Verified score and has autonomously discovered decades-old bugs in OpenBSD and FFmpeg. Project Glasswing establishes a new precedent for cleared AI access restricted to critical infrastructure and financial sectors, while Treasury Secretary Scott Bessent’s intervention underscores the growing speed gap where AI-driven exploits outpace institutional security patching cycles. Gated availability via AWS and Google Cloud suggests that future high-capability models may never see a general public release.

// TAGS

anthropicclaude-mythos-previewproject-glasswingllmsafetyregulation

DISCOVERED

46d ago

2026-04-14

PUBLISHED

46d ago

2026-04-13

RELEVANCE

10/ 10

AUTHOR

GhaithAlbaaj

// KEEP READING

More AI developer news from the feed

EXPLORE FULL FEED

NEWS4m ago

Uber's Claude Code bill tests AI ROI

The video uses Uber’s reported Claude Code spend as a concrete example of the rising tension around agentic coding tools: usage can scale quickly inside engineering teams, but leadership is still struggling to connect that spend to shipped consumer features. It frames Claude Code as genuinely useful, but also as the kind of token-heavy workflow that is easy to adopt and hard to justify when budgets tighten.

RESEARCH7m ago

UserHarness reframes Theory of Mind as user-mind reconstruction

UserHarness is an inference-time framework for Theory-of-Mind tasks that models a user’s partial observations, evolving beliefs, intentions, and actions instead of inferring mental state indirectly. In the paper, the approach is evaluated across five benchmarks and reaches up to 95.94% macro accuracy, with reported gains of more than 15% relative over existing inference methods and about 20% relative over the strongest prompt-only harness.

UPDATE10m ago

Claude Code adds dynamic workflows with Ultracode mode

Anthropic’s latest Claude Code update adds dynamic workflows that let Claude plan work, fan tasks out across parallel subagents, verify results, and return a single coordinated answer. The new `ultracode` setting raises effort automatically and lets Claude decide when to use the workflow mode, targeting large debugging, codebase migrations, security audits, and other long-running engineering jobs.