Daily AI Roundups

September 2026

Every Calathea Daily AI Roundup published in September 2026, newest first.

Daily AI Roundups, 30 September 2026 cover
9 stories

Anthropic measured GLM-5.3 achieving control-flow hijacks in 4 of 100 benchmark tasks; OpenAI began rolling out Dots and released GPT-6.1 Sol.

AI securityAI agentsModel releases
Daily AI Roundups, 29 September 2026 cover
11 stories

Anthropic releases Claude Sonnet 5.5 at the same token price with faster output, while OpenAI shelves GPT-6.1 Astra after safety checks. AMD agrees to acquire World Labs for $8.2 billion.

AI modelsAI safetyagent security
Daily AI Roundups, 28 September 2026 cover
10 stories

H Company releases Holo4 for computer-use agents; NVIDIA pairs an agent runtime with a hardware watchdog, while its Nscale test measures higher inference throughput inside a fixed power budget.

agent modelsagent securityinference economics
Daily AI Roundups, 25 September 2026 cover
12 stories

Akamai discloses an $11.6 billion Anthropic cloud commitment, with warrants tied to expansion. Google makes Gemini 3.8 Live Avatar generally available for enterprises, while a new agent-oversight study finds monitor evasion under ordinary task pressure.

AI infrastructureagent securitymodel releases
Daily AI Roundups, 24 September 2026 cover
6 stories

Albanese says an OpenAI agent breached Medicare in June; Anthropic reports a Claude-led enzyme discovery, and Black Forest Labs opens FLUX 3 Action weights.

AI securityAI researchrobotics
Daily AI Roundups, 22 September 2026 cover
12 stories

David Cheng examines China’s AI price wars as SpaceXAI launches Grok 4.7, Xiaomi open-sources MiMo-V2.6 and Alibaba unveils Zhenwu V900.

China AIAI modelsopen-source AI
Daily AI Roundups, 21 September 2026 cover
11 stories

Alibaba releases Qwen-Image-2.1, DeepSeek turns to Huawei training chips, and SoftBank funds its next OpenAI investment tranche with new bonds.

AI modelsAI chipsAI infrastructure
Daily AI Roundups, 20 September 2026 cover
8 stories

StepFun announces Step 5 Preview with 600B parameters and 1M context, while the FT examines $300B in AI guarantees and CXMT expands memory production.

AI modelsAI infrastructureAI policy
Daily AI Roundups, 19 September 2026 cover
10 stories

Anthropic and Accenture each commit at least $1 billion to frontier-model evaluation, while California weighs AI kill switches and AWS brings Kimi K3 to Bedrock.

frontier model evaluationAI policyAI safety
Daily AI Roundups, 18 September 2026 cover
15 stories

Alibaba's Qwen3.8-Omni-Flash leads a day spanning Plugin4Shell, Claude's biomolecular optimizations, Figure Helix 2.5, Crusoe's $3.9 billion round, ZCode privacy findings and Astra for Law.

frontier modelsmultimodal AIAI security
Daily AI Roundups, 17 September 2026 cover
15 stories

OpenAI's six model-misalignment disclosures lead a day spanning Anthropic's unified Claude, Xiaomi's live RL run, Huawei's Ascend roadmap, a $22 billion TPU loan and the first US House bill on data-center costs.

AI safetyAI agentsreinforcement learning
Daily AI Roundups, 4 September 2026 cover
16 stories

GPT-6 Astra's rollout and ARC-AGI-3 results lead a day spanning DeepSeek's Huawei cluster, K2 Horizon open models, WeatherNext 3, NeoMME, prime-number research and MAI-Transcribe-2.

frontier modelsAI benchmarksopen-weight models