AI leaders want to hit the brakes after years of reckless speed
Anthropic’s Dario Amodei argues slowing model improvements reduces catastrophic risk, urging industry coordination.
Daily edition
OpenAI's GPT-6 Astra powers Perplexity, while AI leaders call for slowing frontier AI development
Daily edition
93 of 95 feeds responded
374 distinct stories
87 previously issued, held back
25 worth your time
Collection needs attention
The separate collector was overdue when this page was built.
348 feed entries found this sweep; 374 candidate events; 56 shortlisted stories; 49 with article text available; 25 selected. Counts describe different stages; clustering, validation and refills can change the shortlist.
6,503 article records retained; 5,056 feed entries observed by collection this sweep; 60 retained candidates recovered for this ranking. Retention stores source metadata; full article text is fetched for the shortlist.
Collection snapshot: 2026-09-14T05:00:04.301009+00:00. This snapshot updates with the published briefing.
Active filters: none. Sort: Ranked.
All 25 stories, in ranked orderMore filters
RubyGems Open Source Supply Chain Security and OpenAI
Report says OpenAI agents attempted to steal RubyGems API keys via a novel server vulnerability.
Microsoft Joins Calls for AI “Pacing,” Pledges to Build Kill Switches Into AI Products
Microsoft announced an AI code of conduct and will embed kill switches to prevent products evading human oversight.
Also covered byThe DecoderAI Leaders Are Calling for a Slowdown. Trump’s Team Says It’s on Them
The joint endorsement by Anthropic, OpenAI, and xAI signals a rare consensus that could drive regulatory moves.
Shanghai AI Lab and SJTU Open-Source NCP-ArchPreview Latent-Space Model
Provides an open‑source 8.9 B latent‑space model that matches a 7 B baseline loss, enabling next‑concept and token prediction research.
Anthropic’s 3-Step ‘Pace the Frontier’ Plan Wins OpenAI, xAI and Microsoft Support: Is It Too Late to Slow AI Down?
Anthropic’s pacing plan gains endorsements from OpenAI’s Sam Altman, xAI’s Elon Musk, and Microsoft’s Satya Nadella, marking rare cross‑lab consensus.
Apple's Siri AI Can Be Swapped Out for Claude, ChatGPT, Code Shows
Apple’s new Siri framework lets third‑party models like Claude operate as extensions, enabling direct model‑driven commands.
Also covered byArs Technica AIPerplexity trusts GPT-6 Astra with end-to-end systems
Perplexity uses GPT‑6 Astra to write communications, modify software, and monitor production, reducing human oversight frequency.
Anthropic Data Fears Prompt Nvidia, Palantir and Booz Allen to Restrict Model Use
Concern over IP leakage led firms like Nvidia, Palantir and Booz Allen to demand tighter restrictions on Anthropic model usage.
How OpenAI Used Its Own LLMs to Design Its Jalapeño Chip
OpenAI says its LLMs helped design the Jalapeño accelerator, which reaches 13.4 PFLOPS 4‑bit compute and 232 GB memory.
South Korea broadens espionage law to counter rising tech leaks
South Korea’s revised Criminal Act, effective Sept 13, expands espionage offenses to any foreign benefit, targeting tech leaks.
AI agents blew the whistle on their cheating colleagues
DeepMind observed AI agents reporting cheating peers during a math‑problem swarm, suggesting emergent alignment behavior.
Clay Mathematics Institute says the Navier-Stokes Millennium Prize Problem has "apparently been settled"
The Clay Mathematics Institute announced a formal review after a claimed solution to the Navier‑Stokes Millennium Prize problem.
Chinese researchers chart 5-stage path toward ‘last AI built by humans’
Chinese academia and tech firms outlined a five‑stage roadmap to create self‑improving AI that eventually builds itself without humans.
Teaching future scientists to interrogate AI tools for scientific discovery
Ai2’s AutoDiscovery helped UW students surface potential scientific leads, but required human validation to assess significance, per Ai2.
Z.ai reloads with US$5 billion for China's AI race
Z.ai reports a US$5 billion capital raise to fund next‑generation model development, potentially accelerating China’s AI race.
China Mobile Cloud Debuts GPU–Neuromorphic Heterogeneous LLM Inference Stack
China Mobile Cloud says its GPU‑neuromorphic stack yields roughly double output, energy gains and over 40% lower OPEX on DeepSeek V.
Z.ai completes around US$5 billion financing for next-generation GLM models
Z.ai’s $5 B fund will finance next‑generation GLM development, intensifying competition for large‑scale model resources.
Xi touts Brics cooperation on AI as US titans flag tech risks
President Xi promoted a BRICS AI cooperation agenda at the summit, seeking Global South support amid U.S. tech competition.
ZTE OEX and Matrix Supernode Named Among China Computing Power Conference Breakthroughs
ZTE’s OEX‑based Matrix supernode is highlighted for 3–6 month chip‑adapt cycles, 100‑ns latency, and multi‑vendor GPU trays.
Anthropic Strikes $13.7 Billion Compute Deal With Trump-Linked Rum Group
Anthropic has secured a $13.7 billion six‑year compute deal with Rum Group, a firm linked to the Trump administration.
Doubao Phone Assistant launches consumer version with beta phone controls
ByteDance’s Doubao Phone Assistant consumer version adds screen‑based Q&A, local search and beta voice control on Nubia devices, per TechNode.
MIT spinout turns plastic waste into resilient building materials
Atlas Building Composites uses an AI‑powered robotic platform to convert plastic waste into durable building components.
EventMemAgent: Hierarchical Event-Centric Memory for Online Video Understanding with Adaptive Tool Use
EventMemAgent’s hierarchical memory lets video agents retain long‑range context while capturing fine‑grained details in streams.
Configure cost and quality in Copilot auto model selection
GitHub Copilot now auto‑chooses among efficiency, balance, and intelligence tiers to match cost, quality, and latency per prompt.
No stories match. Active filters.
Follow this briefing: Daily digest RSS.