friday, june 26, 2026
top 7 trending●previous 24h●generated 40d ago
The Trump administration authorized Anthropic to release Mythos 5 to over 100 US companies and government agencies after a two-week negotiation, de-escalating restrictions on the advanced AI model. The public-facing variant, Fable 5, remains restricted with no rollout timeline.
- Mythos 5 approved for 100+ US organizations including non-American employees
- Commerce Secretary Howard Lutnick signed authorization letter to Anthropic co-founder Tom Brown on June 26
- Fable 5 (public-facing Mythos-class model) still in limbo with no release agreement
- Two-week negotiation period between Trump administration and Anthropic concluded June 27
Claude Opus 4.7 completed a complex end-to-end software engineering task in 14 hours that would typically require 2–17 weeks of human work, costing $251. The experiment demonstrates rapid improvement in AI coding capabilities, though models remain imperfect.
- Opus 4.7 built a full software package in 14 hours versus 2–17 weeks for human engineers
- Total cost for the task was $251
- Experiment conducted by @emollick on 2026-06-26
- Models show fast improvement but still have limitations
- Opus 4.8 and GLM-5.2 exhibit notably different personalities in reasoning tasks
The Trump administration requested that OpenAI stagger the release of GPT-5.6, and OpenAI agreed to limit the model's rollout. The request came after direct communication between the White House and OpenAI in late June 2026.
- Trump administration made formal request for staggered GPT-5.6 release on June 25-26, 2026
- OpenAI agreed to limit and delay the next model's rollout following White House call
- Staggered release approach replaces standard full deployment strategy
- Request reflects government interest in controlling pace of advanced AI model availability
OpenAI released GPT-5.6 as a limited preview to government-vetted users after the Trump administration requested delayed rollout of the model, which features three tiers (Sol, Terra, Luna) with strengths in coding, cybersecurity, and biology. The company publicly objected to the restriction, stating such government-controlled access processes should not become standard practice.
- GPT-5.6 suite includes Sol (flagship), Terra (medium-tier), and Luna (fast/affordable) models
- Sol priced at $5 per million input tokens, $30 per million output tokens
- Initial rollout limited to ~20 government-vetted trusted partners before wider release
- OpenAI CEO Sam Altman told employees of government request for restricted access
- Trump administration also recently ordered Anthropic to restrict access to Mythos/Fable 5 model
The Linux Foundation launched Akrites, a new initiative designed to protect open-source software from AI-enabled security exploits and attacks. The project addresses growing concerns about automated vulnerability discovery and exploitation targeting FOSS projects.
- Linux Foundation announced Akrites on June 26, 2026
- Akrites specifically targets AI-enabled exploits against open-source software
- Initiative aims to defend FOSS projects from automated attack vectors
- Addresses emerging threat of AI-powered vulnerability discovery in open-source ecosystems
Anthropic accused Alibaba of conducting the largest-ever Claude distillation attack, involving 28.8 million fraudulent API calls to illicitly extract and replicate Claude's capabilities. The company characterized the campaign as 'brazen' and unprecedented in scale.
- 28.8 million fraudulent API calls used to extract Claude model capabilities
- Alibaba allegedly ran coordinated distillation campaign to replicate Anthropic's AI
- Anthropic described the attack as the largest of its kind on record
- Illicit access involved systematic extraction of Claude's underlying model weights and behaviors
OpenAI previewed GPT-5.6, a tiered model family consisting of Sol, Terra, and Luna variants with new reasoning modes including max and ultra options. Sol, the lead model, shows strengthened performance in coding, science, and cybersecurity with an advanced safety stack, available in limited preview.
- GPT-5.6 released in three tiers: Sol, Terra, and Luna models
- New reasoning modes: max and ultra options for enhanced problem-solving
- Sol variant emphasizes coding, science, and cybersecurity capabilities
- Advanced safety stack integrated into the model family
- Limited access preview launched June 26, 2026