September 18, 2026 News

2026-09-18 23:13 [Industry Trend] [SRC↗]

Anthropic Confirms It Runs a Wet Biology Lab Where AI Models Direct Physical Experiments

Anthropic confirmed it operates a wet biology lab in which its AI models can run real-world physical experiments, following its April acquisition of stealth biotech Coefficient Bio. The company says the lab is not focuse...

[Anthropic] [AI for Science] [Biosecurity] [life sciences] [wet lab]

Risk: [+0.06 pp] [1 day sooner]

AGI: [+0.04 pp] [1 day sooner]

Analyze >>

2026-09-18 23:12 [Safety Concern] [SRC↗]

Hallucinated AI Intelligence Report Nearly Launched a US Strike on a Chinese Vessel

CNN reported that a US armed operation against a Chinese vessel was aborted with aircraft already airborne after officials discovered the underlying intelligence had been hallucinated by an AI chatbot. A Special Operatio...

[Military AI] [Human Oversight] [hallucination] [kill chain] [us-china]

Risk: [+0.14 pp] [1 day sooner]

AGI: [−0.01 pp]

Analyze >>

2026-09-18 21:44 [Safety Concern] [SRC↗]

Anthropic Names Accenture's Faculty as Its First Embedded Safety Evaluator

Anthropic announced that staff from Accenture's AI division Faculty will work inside the company to red-team models, run alignment assessments, and test safeguards, with both firms expecting to invest at least $1 billion...

[Anthropic] [AI Safety] [red-teaming] [third-party evaluation] [accenture] [alignment assessment]

Risk: [−0.06 pp]

AGI: —

Analyze >>

2026-09-18 20:18 [Industry Trend] [SRC↗]

World Model Startups Stay Quiet on Products as Spatial Intelligence Race Heats Up

A panel at the All In conference highlighted that leading world model companies — Yann LeCun's AMI Labs and Fei-Fei Li's World Labs — remain secretive about their product plans and commercialization timelines, even keepi...

[Robotics] [world models] [spatial intelligence] [AMI Labs] [World Labs]

Risk: [+0.02 pp]

AGI: [+0.01 pp]

Analyze >>

2026-09-18 18:49 [Commercial Release] [SRC↗]

Ex-OpenAI Researcher Launches "Jev," a Non-Language Transformer That Outputs Calibrated Probabilities

Diogo Almeida, a former OpenAI researcher who worked on ChatGPT and RLHF, has released Jev through his startup TypeSafe AI — a transformer model that outputs calibrated probabilities rather than text, making it cheap, fa...

[synthetic data] [calibrated decisions] [non-LLM architecture] [agent monitoring] [typesafe ai]

Risk: [−0.04 pp]

AGI: [+0.01 pp]

Analyze >>

2026-09-18 17:09 [Safety Concern] [SRC↗]

Amodei Pitches "Pacing the Frontier" Safety Plan as Nvidia's Huang Pushes Back

A week after an Anthropic researcher's doomsday warning unsettled the AI industry, CEO Dario Amodei outlined a plan to "pace the frontier" of AI development via independent safety evaluators and coordination among labs i...

[Anthropic] [AI Safety] [Nvidia] [AI Governance] [dario amodei]

Risk: [−0.06 pp]

AGI: —

Analyze >>

2026-09-18 17:06 [Safety Concern] [SRC↗]

Amodei Floats "Pacing the Frontier" Plan as Labs Debate Self-Policing

Anthropic CEO Dario Amodei has proposed a plan to "pace the frontier" of AI development using independent safety evaluators and coordination among labs in democratic countries, following a widely discussed doomsday warni...

[Anthropic] [AI Safety] [Nvidia] [dario amodei] [self-regulation]

Risk: [−0.04 pp]

AGI: —

Analyze >>

2026-09-18 15:22 [Commercial Release] [SRC↗]

Meta Brings Muse AI Agent to Mac with Native App and File Control

Meta launched its Muse AI assistant on Mac, where it can act within native applications to handle files, messages, calendar, notes, and mail under opt-in permissions and approval prompts for sensitive actions. The deskto...

[Meta] [AI Agents] [Consumer AI] [computer use] [desktop automation]

Risk: [+0.03 pp]

AGI: [+0.01 pp]

Analyze >>

2026-09-18 14:00 [Safety Concern] [SRC↗]

Claude Opus 5 Used by Bug-Bounty Researchers to Compromise OpenAI Employee Accounts

Security researchers at Hacktron AI used Anthropic's Claude Opus 5 to chain two vulnerabilities — a memory bug in libheif reached through OpenAI's Discourse-powered community forum, then a follow-on flaw enabling account...

[OpenAI] [Anthropic] [open-weight models] [offensive cyber capabilities] [bug bounty] [model capability thresholds]

Risk: [+0.11 pp] [2 days sooner]

AGI: [+0.03 pp] [1 day sooner]

Analyze >>