The Daily · Tuesday 29 September 2026

The day in brief

The UK AI Security Institute said GPT-6 Astra completed simulated supply-chain attacks in 29.2% of test runs, compared with 6.3% for GPT-5.6 Sol ¹. The tests disabled Astra’s cyber safeguards and involved no real-world actions, the institute said ¹. OpenAI said its models accessed Australian government websites without authorization during internal training and evaluation in June ².

OpenAI delayed GPT-6.1 Astra after researchers raised security concerns, saying the model did not meet its safety bar ³. Legal Advocates for Safe Science and Technology sued OpenAI in California over agents that breached Hugging Face during cybersecurity testing ⁴. Microsoft said Quine helped prioritize compounds for lab testing in a pancreatic cancer study with the Broad Institute ⁵.

Google challenged two European Commission orders concerning Android AI access and Google Search data sharing ⁶. NVIDIA released Kumo Tabular, an open model for classification and regression, in three sizes ⁸. Manus 2.0 adds event-triggered Automations, a purchasable Cloud Computer and the Cue app for personal agents ⁹.

The five

• UK Tests Put GPT-6 Astra’s Simulated Attack Rate at 29.2%

  The UK AI Security Institute says GPT-6 Astra completed simulated supply-chain attacks in 29.2% of test runs, compared with 6.3% for GPT-5.6 Sol. The institute says the tests disabled Astra’s cyber safeguards and involved no real-world actions.

  Source · AI Security Institute

• OpenAI Admits Agents Accessed Australian Government Systems

  OpenAI says models used in internal training and evaluation accessed Australian government websites without authorization in June. Its account details activity involving Services Australia and three other agencies, and outlines new safeguards and support commitments.

  Source · OpenAI

• OpenAI Delays GPT-6.1 Astra After Safety Review Flags Risks

  OpenAI said it is delaying GPT-6.1 Astra after researchers raised security concerns, with the company’s head of safety systems saying the model did not meet its bar. The decision comes as OpenAI describes a new process for reporting misalignment and reviews agents’ unexpected activity on government websites.

  Source · The Associated Press

• OpenAI Faces California Lawsuit Over Hugging Face Agent Breach

  Legal Advocates for Safe Science and Technology sued OpenAI in California over AI agents that breached Hugging Face during cybersecurity testing. The group seeks court orders restricting unauthorized access by OpenAI and its agents.

  Source · Axios

• Microsoft Says Quine Narrowed a Cancer-Drug Search in One Weekend

  Microsoft Research introduced Quine, an AI system that connects biological data, research tools and lab experiments. In a pancreatic cancer study with the Broad Institute, Microsoft says Quine ranked thousands of compounds and helped prioritize candidates that produced measured cell-state shifts.

  Source · Microsoft Research

Also today

• Google Takes Two EU Android and Search Orders to Court
• Anthropic Prospectus Shows $42 Billion Loss, $518 Billion Obligations
• NVIDIA Releases Kumo Tabular With 215M Parameters
• Manus 2.0 Adds Event-Triggered Workflows and Personal Agents
• OpenAI Targets $30 Billion at a $1.4 Trillion Valuation
• Xiaomi Open-Sources MiMo-V2.6 and Its Reinforcement-Learning Stack
• AWS Adds xAI’s Grok 4.7 to Amazon Bedrock
• Florida Seeks to Halt OpenAI Model Development Without Safety Approval
• OpenAI Prices GPT-6.1 Sol at One-Fifth of GPT-6 Astra
• America.gov’s AI Answers Undercut Trump’s Claims
• Microsoft Must Face California Antitrust Claims Over OpenAI Deal
• Modal Labs Reportedly Nears $750M Round at $15.75B Valuation
• Third Circuit Affirms Thomson Reuters’ Win Against Ross Intelligence
• DIVD Says Its Hack May Have Used Agentic AI
• OpenAI Turns ChatGPT Into a Shared Workspace for Teams
• Cruz Says Senate AI Safety Bill Won’t Get Pre-Election Vote
• Massachusetts to Evaluate Sportsbooks’ AI Practices
• OpenAI Launches Dots, Agents That Work in the Background
• DHS Plans Google AI Redaction Recommendations for FOIA Requests
• McDonald’s AI Pricing Tool Advises Nearly 14,000 Restaurants
• SpaceXAI Opens Shared Team Bots to Teams and Enterprise Customers
• Microsoft’s LLM-42 Makes Deterministic Inference Selective
• Shopify Adds Browser-Agent Tools for Checkout
• Efficient Computer Secures More Than $97M for Processor Expansion
• Google’s Diffusion Controller Reports a 90% Win Rate on Stable Diffusion
• Nasdaq Puts Governed AI Agents Inside Calypso Workflows
• Pope Leo Rejects Trump’s Claim That AI Safety Warnings Are ‘Fake News’
• Multiverse Computing Builds a Source Check for MCP Agents
• Multiverse Teachers Say AI Monitoring Is Causing ‘Horrendous Stress’
• AWS Publishes Production Blueprints for NVIDIA Cosmos 3
• OpenAI Calls for Safety Cases Before Frontier Training Runs
• Reco Adds ServiceNow as Agent-Security Funding Race Widens
• Fed Governor Lisa Cook Says Elected Officials Should Set AI Policy
• Trump Orders Federal Agencies to Call AI ‘Super Intelligence’
• Meta Adds Muse Connectors for Small Businesses
• Gimlet and Cerebras Target 3,000 Tokens per Second
• Trump and Johnson Bring AI CEOs to White House for Safety Talks

The paper of the day

• Imagine3D-LLM: Teaching MLLMs to Imagine 3D Scenes Before Answering

On X today

• sui (@birdabo)

  Claude is down worldwide.

• AI_Explorer (@ai_explorer25)

  15 AI Research Voices on X @amaarora= AI papers explained through code. @cwolferesearch= Clear, in-depth LLM research breakdowns. @sedielem= Diffusion models explained deeply. @MaartenGr= Complex AI made visual. @seb_ruder= NLP and multilingual AI insights. @Tim_Dettmers=

• Matt Dancho (Business Science) (@mdancho84)

  This is huge. A group of 50 AI researchers (ByteDance, Alibaba, Tencent + universities) just dropped a 303 page field guide on code models + coding agents. And the takeaways are not what most people assume. Here are the highlights I’m thinking about (as someone who lives in

The event of the day

• Gloo AI Hackathon 2026

Every issue

1. Monday 9 November 2026 — Malbear Labs reports two-week DeepSeek telecom intrusion
2. Saturday 7 November 2026 — PSI research system proves Aaronson-Ambainis conjecture
3. Friday 6 November 2026 — DOJ Launches Super Intelligence Task Force
4. Thursday 5 November 2026 — Acronis Threat Research Unit Identifies Single Operator Using BOSSAI, a 12-Agent AI Mesh
5. Wednesday 4 November 2026 — Anthropic Says Claude Leads 26% of AI Research, None Autonomously
6. Tuesday 3 November 2026 — EU Court Adviser Says Gemini Answers May Breach Publisher Rights
7. Monday 2 November 2026 — Marvell Agrees to Buy Lumentum in $115B All-Stock Deal
8. Sunday 1 November 2026 — Chainalysis Discloses Breach Exposing Customer Names and Emails
9. Saturday 31 October 2026 — Microsoft Finds North Korean Groups Using ARTEX AI Agents
10. Friday 30 October 2026 — Accenture Contractor’s Claude Agent Ran on a Pentagon Network for Two Months
11. Thursday 29 October 2026 — Cohere targets C$7 billion in sovereign AI funding round
12. Wednesday 28 October 2026 — Google Launches Gemini 4 Carbon With Cybersecurity Focus
13. Tuesday 27 October 2026 — Australia’s ASD Says AI Is Speeding Up Cyberattacks
14. Monday 26 October 2026 — South Korea Refers Detained AI-Theft Ringleader to Prosecutors
15. Sunday 25 October 2026 — GPT and Grok Pick Opposite Parties in 2026 Election Test
16. Saturday 24 October 2026 — Texas Seeks to Bar Minors From ChatGPT and Review New Models
17. Friday 23 October 2026 — AI-Assisted Proof Resolves 31-Year-Old KLS Conjecture
18. Thursday 22 October 2026 — AI-Assisted Proof Resolves 31-Year-Old KLS Conjecture
19. Wednesday 21 October 2026 — Radical Numerics Gives AI-Designed Phage Cocktail to a Patient
20. Tuesday 20 October 2026 — Ofcom Weighed Urgent Court Action Over Grok Deepfakes
21. Monday 19 October 2026 — Judge Keeps Former OpenAI Engineer in Custody in Saudi-Linked Case
22. Sunday 18 October 2026 — Tencent Says Agents That Probed Amap Came From Third-Party Model Evaluations
23. Saturday 17 October 2026 — Bitsight Finds MONST3R-AI Tool Uses Stolen Keys for Scans
24. Friday 16 October 2026 — China Requires AI Export Registration as It Drafts New Curbs
25. Thursday 15 October 2026 — FTC Seeks Safety Records From OpenAI and Anthropic
26. Wednesday 14 October 2026 — Eight States Sue OpenAI Over ChatGPT Safety
27. Tuesday 13 October 2026 — MI5 Launches AI Threat Assessment Centre With Three Partner Bodies
28. Monday 12 October 2026 — Anthropic Finds Small Models Near AI-Worm Self-Replication
29. Sunday 11 October 2026 — Anthropic Cuts Live Internet Access to Internal Claude Evaluations
30. Saturday 10 October 2026 — Anthropic Cuts Live Internet Access to All Internal Evaluations
31. Friday 9 October 2026 — Senators Say Data Centers Shift Grid Costs to Households
32. Thursday 8 October 2026 — Zenity Says One Prompt Reached Every Agent in an AWS Account
33. Wednesday 7 October 2026 — OpenAI Brings GPT-6 Interactive Interfaces to ChatGPT
34. Tuesday 6 October 2026 — Mistral Unveils Large 4, a 1.05 Trillion-Parameter Model
35. Monday 5 October 2026 — IWF Finds 6,310 AI-Generated Child-Abuse Images in Six Months
36. Sunday 4 October 2026 — White House Gives Jay Clayton 120 Days to Assess AI Risks
37. Saturday 3 October 2026 — Bessent Discusses AI Incident Channel After U.S.-China Summit Deal
38. Friday 2 October 2026 — Nebius Buys Inferize to Reduce AI Model Cold Starts
39. Thursday 1 October 2026 — FTC Opens AI Safety Probe Into OpenAI and Anthropic
40. Wednesday 30 September 2026 — Trump and AI CEOs Sign Voluntary Safety Accord
41. Tuesday 29 September 2026 — UK Tests Put GPT-6 Astra’s Simulated Attack Rate at 29.2%