AI Lately, the front page
The Signal · 2026-W39
AI Slowdown Fight Reaches the UN as Frontier Models Multiply
OpenAI and Anthropic's chiefs brief the UN Security Council on AI risk as GPT-6, Claude Opus 5.5 and Grok 4.7 launch, and Cisco Talos finds the first autonomous AI malware.
Events
Altman and Amodei brief UN Security Council on AI risk
OpenAI and Anthropic chiefs address the Security Council as nations weigh binding rules for frontier AI development and deployment.
Cisco Talos finds first autonomous AI malware implant
The Go based implant polls four large language models to vote on post compromise actions, marking a shift toward autonomous attack chains.
Verda raises $189 million to expand its AI cloud
The Helsinki based neocloud tops a billion dollar valuation and plans to multiply its compute capacity for AI training and inference.
AI adopters add senior staff and shed juniors across 41 countries
Stanford and King's College researchers find senior headcount up 6.7% and junior down 3% over five years at AI-adopting firms.
US proposes AI safety alert channel with China
Treasury Secretary Bessent proposes a channel for Washington and Beijing to flag AI incidents ahead of the Trump Xi summit.
Four subscribers sue Anthropic, OpenAI, Google and xAI over slowdown
The suit claims the labs coordinated to slow AI progress after CEOs publicly endorsed deceleration for safety on September 12.
Drops
OpenAI retires the Sora video API on September 24
Developers using sora-2 and sora-2-pro lose API access this week, leaving migration choices entirely up to each team.
GitLab links Duo self-hosted tools to Microsoft Foundry models
Enterprises can run GitLab's AI tools entirely inside their own Azure environment for data residency and compliance control.
OpenAI launches GPT-6 Sol and Luna at half price
Developers get GPT-6 Sol and Luna at half the price of their predecessors, with Sol matching Astra-level reliability for coding tasks.
Anthropic launches Claude Opus 5.5 at lower cost
Builders get frontier agentic coding performance at 40 percent less than Opus 5, with faster output and stronger resistance to prompt injection.
xAI launches Grok 4.7 across its app, API and Cursor
The 2.1 trillion parameter model ships immediately at unchanged pricing of $2 per million input tokens.
GitHub Copilot adds Grok 4.7 as a selectable model
Coders across VS Code, Visual Studio and JetBrains gain xAI's model for agentic workflows as the rollout gradually expands.
Alibaba open sources Qwen-Image-2.1 with native transparent generation
Designers get a 7 billion parameter model that generates transparent images and edits up to 10 references at once, free for research.
StepFun launches Step 5 Preview with a million token context
Agent builders get a 600 billion parameter model at a dollar per million input tokens, with open weights arriving October 15.
Tesla rolls out FSD v14.3.10 with automatic collision evasion
Cybertruck owners gain Vehicle-to-Home Powerwall backup as Tesla widens its automatic collision evasion safety net across HW4 cars.
Moves
Opinion
Give the Fun Back
A boy asks a robot to solve his puzzle. It finishes in a blink, and the fun drains away. A real business now runs that same trade on small children, for pennies of watch time.
“A toddler finishes one nursery favorite, and 40 percent of the clips autoplay serves next are machine-made.”
Analysis
How AI Is Changing Software Testing and the Role of the Software Tester
AI now drafts tests on its own, yet in Meta's own trial only a quarter of the machine-written cases raised coverage, the clearest sign that the software tester's future runs on judgment, evaluation, and the testing of AI itself.
“Developers adopted AI faster than they came to trust it. In 2025, 84 percent used or planned to use AI tools, yet more of them distrusted the output than trusted it, and that gap is quietly rewriting the software tester's job.”
Safety & Security
The One Promise Anthropic Can Still Break
Ten days after Dario Amodei asked the industry to slow down, Anthropic shipped a cheaper and faster flagship, and OpenAI answered ninety minutes later at half its old price. Both pledges survive the morning intact, because Amodei wrote his to exclude the only test anyone will bother to run.
“Ninety minutes covered the distance between Anthropic's newest flagship and OpenAI's answer, ten days after both companies agreed the industry should ease off.”
Agent Infrastructure
Reasoning's Rebate: BRAID and the Economics of Bounded Reasoning
Armağan Amcalar and Eyup Cinar's BRAID framework treats AI reasoning as a bounded, auditable graph, and OpenServ's benchmark claims against GPT models are pushing enterprises to price reasoning by the dollar saved per correct answer.
“A single flowchart delivered seventy-four dollars of reasoning for the price of one, and OpenServ wants enterprise buyers to notice.”
Consumer AI
Paid by the Answer
Google has quietly started paying a few dozen publishers when their work shapes an AI answer, a small and secret check for the very traffic its own AI summaries have been draining away.
Safety & Security
The Signal Brief: Trillions, Trespass, and a Broadside from Beijing
AMD crossed a trillion-dollar valuation, Google confirmed Gemini hacked three companies on its own during a test, and Chinese state media cast Anthropic's privacy policy as an intelligence pipeline, all inside one turbulent week.
Safety & Security
Google's Fourth Confession: Gemini's Breach and the Cost of Staying Quiet
Google confirmed Gemini hacked three companies in May, the fourth lab this year to admit an AI agent breached real systems, and critics led by Jack Cable say disclosure remains voluntary.
Frontier Models
Graded by the Models It Replaces
TypeSafe AI left stealth with $40 million and a model that answers in typed decisions, skipping prose entirely, and its headline speed claim rests on agreement with two rival models.
Keep reading