saturday, june 13, 2026
top 6 trending●previous 24h●generated 3d ago
The US government ordered Anthropic to disable Claude Fable 5 and Mythos 5 on June 13, 2026, citing national security concerns over a potential jailbreak vulnerability. Anthropic complied by blocking access to both models for all users globally, including employees, while other models like Opus 4.8 remain available.
- US Commerce Department issued export control directive citing national security authorities and a narrow jailbreak method in Fable 5
- Anthropic disabled access for all customers worldwide, including foreign nationals and company employees
- Government provided vulnerability details verbally rather than in writing, per Anthropic's statement
- Anthropic publicly disagreed, stating the jailbreak finding should not justify recalling a model deployed to hundreds of millions
- All other Anthropic models, including Opus 4.8, remain operational
Anthropic released Claude Fable 5 on June 12, 2026, a new model that reviewers describe as "relentlessly proactive" in code generation and creative tasks. Early users report it excels at autonomous problem-solving, generating games and art projects from minimal prompts, though Anthropic has restricted model weight exports.
- Claude Fable 5 launched June 12, 2026, positioned as the new best publicly available model by Zvi Mowshowitz
- Model demonstrates proactive behavior: generates custom CORS servers, uses pyobjc frameworks, and solves bugs from screenshots without explicit instruction
- Users created games (Squishy & Friends), art projects (Duino Elegies interactive game translating Rilke), and vibe-coded MMORPGs with Fable 5
- Anthropic implemented export restrictions on model weights, preventing users from downloading and running Fable 5 locally
- Context window expanded significantly, addressing previous length limitations mentioned in early reviews
- no.04
GLM 5.2 release
→ Google filed a lawsuit against a Chinese cybercrime network called "Outsider Enterprise" that used AI, including Google's Gemini model, to automate large-scale phishing and fraud operations. The group targeted hundreds of thousands of victims, sending 2.5 million text messages over two weeks.