The Daily · Friday 2 October 2026

The day in brief

Nebius said Inferize’s technology and team had joined Token Factory, where integration is starting ¹. The technology is designed to reduce model cold starts and GPU capacity tied up during startup ¹. Black Forest Labs launched FLUX 3 Image with bounding-box editing, native 2K and 4K output, and support for up to 10 reference images ².

Modal made multi-node GPU clusters generally available to all workspaces, with gang scheduling, per-second billing and RDMA networking ³. GitLab patched a critical flaw that could let an authenticated user with Duo Agent Platform access execute commands on a self-hosted AI Gateway ⁴. Google said its new system cut one Gboard model’s training time from about two months to three weeks ⁶.

SoftBank completed the final $10 billion tranche of its follow-on OpenAI investment, bringing the commitment to $30 billion ¹⁰. Perplexity released an Apache 2.0-licensed 27B decision model designed to return choices and probabilities for classification and routing ⁸. Amazon Web Services CEO Matt Garman said more than 100 data-center moratoriums are under consideration in the United States ⁷.

The five

• Nebius Buys Inferize to Reduce AI Model Cold Starts

  Nebius says Inferize’s technology is designed to reduce the time and GPU capacity tied up when large AI models start or scale. The company announced on October 1, 2026, that Inferize’s technology and team had joined Token Factory, with technology integration beginning there.

  Source · Nebius

• Black Forest Labs Launches FLUX 3 Image With Box-Level Editing

  Black Forest Labs launched FLUX 3 Image with targeted image editing, layouts controlled by bounding boxes and outputs up to 4K. The company is offering 50% off API use through October 8 and says an open-weights version will follow in the coming weeks.

  Source · Black Forest Labs on LinkedIn

• Modal Makes Multi-Node GPU Clusters Generally Available

  Modal opened its multi-node GPU clusters to all workspaces on October 1, 2026, with gang scheduling, per-second billing and RDMA networking. The service uses a Python decorator to request coordinated GPU nodes, but requires full-GPU nodes and remains bounded by each workspace’s plan limits.

  Source · Modal

• GitLab Patches CVSS 9.9 RCE in Self-Hosted AI Gateway

  GitLab released three patched AI Gateway versions to address a critical vulnerability that could let an authenticated user with Duo Agent Platform access execute commands on a self-hosted gateway. Customers using GitLab-hosted gateways are already protected, while operators of affected self-hosted installations are urged to upgrade.

  Source · GitLab Docs

• Study Finds Test-Selected Baselines Make Safety Routing Look Worse

  A paper argues that choosing a single-model comparator using evaluation labels favors that comparator and makes safety routing appear less effective. The selection cost is larger when evaluation categories or task suites are held out than under random splits.

  Source · arXiv

Also today

• Google Moves Gboard Training Into Verifiable Server-Side Enclaves
• Amazon Says 100 Data-Center Moratoriums Could Hurt U.S. AI Race
• Perplexity Releases Apache-Licensed 27B Decision Model
• Argo-Bench Finds Data Agents Miss the Decision, Not Just the SQL
• SoftBank Completes Its Final $10 Billion OpenAI Investment
• Trillium Labs Puts Frontier AI Training Research in Public
• Supabase Acquires Turso to Give AI Agents Their Own Databases
• Anthropic Pledges $100 Million to Train 10,000 AI Engineers
• Tencent Reportedly Leases Access to 100,000 GPUs via Oracle
• ChatGPT’s Mac App Flaw Let Local Code Reach Chats and Browser Sessions
• New Mexico Proposal Puts Frontier AI Risk Assessments Under State Audit
• Dynatrace Completes Its $915 Million Arize Acquisition
• IBM Makes Bob Self-Hostable for Sensitive Software
• Cloudflare Releases Clef Decision Models and RL Fine-Tuning Service
• Arizona Appeals Court Tosses Sentence Tied to AI Victim Video
• Apple Adds macOS Controls for Full Disk Access After Muse Dispute
• Amazon Discusses $8 Billion Nvidia Chip Financing Structure, FT Reports
• DOJ Arrests Greg Lui in Alleged $300 Million Server Scheme
• Maryland Finds 91% of Surveyed Businesses Use AI, Mostly in Basic Ways
• A Mixed AI Review Caught More Planted Errors in One Test
• Microsoft Sees AI Compressing Cyberattack Steps to Seconds
• Cantina Releases Apex Flash-1, a 321B Open-Weight Security Model
• MitPro Raises Pathologist Agreement Across 385 Tumor Slides
• SourceLearn Builds Persistent Models of Documents, Code and APIs
• David AI Finds a Sevenfold ASR Accuracy Gap Across Languages
• Ai2 Releases AstaBrief, an 8B Model for Faster Scientific Reports
• NVIDIA Adds $4,999 64GB DGX Spark for Local Agents
• Pi 1.0 Cuts Codemode Prompt Use by About 40%
• Suno Adds Speech Model That Pairs Spoken Voice With Music
• MIT’s InstructMesh Lets Users Repair AI-Generated 3D Models
• Apple Paper Finds Confidence Rankings Can Distort Diffusion Outputs
• Meta Opens Muse to ESP32 and Linux Gadgets
• Ethereum Foundation Launches zkAPI for Private AI Payments
• Arceus Raises $17 Million to Build an AI-Enabled Law Firm
• Bull Doubles Production Capacity at Its €80M Angers Factory
• Applied Digital Brings 75 MW Online at Polaris Forge 1
• Study Finds Patients Penalize Doctors Who Disclose AI Use
• South Korea Orders Bank Audits as Police Probe AI-Suspected Hacks
• Mercor Finds Claude Outperforms CPAs on Four Close Tasks
• Wes Moore Recruits Governors for a Bipartisan AI Framework
• NVIDIA Puts OpenAI’s Astra Ultrafast on Blackwell GPUs
• Sony Brings QSSR AI Upscaling to the Standard PS5
• arXiv Limits Submitters to Two Papers a Month
• Anthropic Seeks Conditional Copyright Approval in Australia
• Ideogram 4.5 Takes Aim at Drift in Repeated Image Edits
• Anthropic Adds TypeScript Mods to Claude Code
• Anthropic Finds Robots Can Do 74% of Physical Tasks, but Cost Too Much
• Inworld Buys Ultravox to Add Voice Agents to Its Speech Stack
• Connecticut’s AI and Privacy Rules Take Effect as 14 Companies Sue
• Apple Researchers Find Language Separation Helps Speech Models
• ChatGPT Hallucinations Put Customers at Odds With Service Workers
• Microsoft Study Finds Seniority, Not AI Disclosure, Skews Code Reviews
• Bloomberg Completes Its Acquisition of Canoe Intelligence
• Palantir Scheduling Tools Draw Error Complaints From Health Workers
• ServiceNow’s AutoSynthData Turns Agent Failures Into Training Tasks
• DigitalOcean Prices Agent Droplets at $50 and $200 a Month
• Anthropic Urges Australia to Allow AI Training With Opt-Outs

The paper of the day

• KaliBench: A Fine-Grained Benchmark for Cybersecurity Tool Use on Kali Linux with Runtime-Free Verifiable Rewards

On X today

• Andrej Karpathy (@karpathy)

  Cool eval. Simply ask an LLM “Land or Water?” and give it a latitude and longitude coordinate as text. Ask 16,200 times, plot as image. The models know. From compressing the internet.

• Max Simchowitz (@max_simchowitz)

  There seems to be fear reasoning models will trivialize research on robot learning architectures, algorithms, etc... I disagree. Innovation comes from constraints, and robotics imposes a ton of then. Our new paper, led by @DashoraNitish , suggests a really promising new

The update of the day

• Stable Diffusion

Every issue

1. Sunday 18 October 2026 — Tencent Says Agents That Probed Amap Came From Third-Party Model Evaluations
2. Saturday 17 October 2026 — Bitsight Finds MONST3R-AI Tool Uses Stolen Keys for Scans
3. Friday 16 October 2026 — China Requires AI Export Registration as It Drafts New Curbs
4. Thursday 15 October 2026 — FTC Seeks Safety Records From OpenAI and Anthropic
5. Wednesday 14 October 2026 — U.S. Regulators Ask Banks How They Would Shut Down AI Agents
6. Tuesday 13 October 2026 — MI5 Launches AI Threat Assessment Centre With Three Partner Bodies
7. Monday 12 October 2026 — Anthropic Finds Small Models Near AI-Worm Self-Replication
8. Sunday 11 October 2026 — OpenAI’s Hugging Face Hack Exposed a Failure in Its Test Sandbox
9. Saturday 10 October 2026 — Anthropic Cuts Live Internet Access to All Internal Evaluations
10. Friday 9 October 2026 — Senators Say Data Centers Shift Grid Costs to Households
11. Thursday 8 October 2026 — Zenity Says One Prompt Reached Every Agent in an AWS Account
12. Wednesday 7 October 2026 — OpenAI Brings GPT-6 Interactive Interfaces to ChatGPT
13. Tuesday 6 October 2026 — Mistral Unveils Large 4, a 1.05 Trillion-Parameter Model
14. Monday 5 October 2026 — IWF Finds 6,310 AI-Generated Child-Abuse Images in Six Months
15. Sunday 4 October 2026 — White House Gives Jay Clayton 120 Days to Assess AI Risks
16. Saturday 3 October 2026 — Bessent Discusses AI Incident Channel After U.S.-China Summit Deal
17. Friday 2 October 2026 — Nebius Buys Inferize to Reduce AI Model Cold Starts
18. Thursday 1 October 2026 — FTC Opens AI Safety Probe Into OpenAI and Anthropic
19. Wednesday 30 September 2026 — Trump and AI CEOs Sign Voluntary Safety Accord
20. Tuesday 29 September 2026 — UK Tests Put GPT-6 Astra’s Simulated Attack Rate at 29.2%