AI
Artificial intelligence developments, regulation, and industry
US forces Anthropic model recall as open-source GLM 5.2 launches
- Frontier AI Intervention: The US government gave Anthropic 90 minutes to recall its newest frontier AI model following executive orders establishing federal review over advanced systems. Anthropic co-founder Jack Clark subsequently proposed an industry-wide pause on frontier AI development, citing the rapid emergence of models capable of designing themselves.
- Open-Source Coding Milestone: Z.ai released GLM 5.2 under a permissive MIT license, delivering a 744-billion parameter mixture-of-experts model with a 1-million-token context window. The model scores 62.1 on SWE-Bench Pro, narrowing the capability gap with proprietary leaders like Claude Opus 4.8 and expanding sovereign technological capacity for developers across the Arab world and the Global South.
- Context Window Breakthrough: AI startup Subquadratic introduced a novel architectural model capable of processing 12 million tokens simultaneously while requiring fundamentally less compute than traditional attention mechanisms. The breakthrough allows systems to ingest massive codebases or entire legal libraries natively without relying on retrieval-augmented generation or vector databases.
- Environmental Accountability: UN Secretary-General António Guterres launched the AI Environmental Transparency Initiative at London Climate Action Week, demanding mandatory disclosures of data centre carbon, water, and land footprints. Guterres warned against shifting infrastructure burdens onto vulnerable regions and urged AI operators to commit to fully renewable energy sources by 2030.
- Benchmark Milestones: DeepSeek V4 has achieved a score of 80.6 on SWE-Bench Verified with a native 1-million-token window, demonstrating exceptional long-horizon agentic capability. The advancement underscores a decisive structural shift where open-weight models now consistently perform within single digits of closed-source frontier architectures.