The Signal — Oct 12, 2026

▲ Important

Claude Haiku 5.5 cuts costs 75%
Anthropic's cheapest model is priced ~75% below Haiku 4.5 for high-volume work like summaries and classification, plus a 50% cut to Sonnet 5.5 cache-read prices. Early testers include Atlassian, HubSpot and Notion.
The AI Report · Oct 11

Google brings agentic AI to Gemini, starting with businesses
Gemini becomes a unified agent with its own Workspace account and audit trail, able to hand off to subagents and use multiple models (including Claude). Enterprise customers get it first.
TechCrunch · Oct 8

Claude Dashboards and Motion launch with Adobe
Two new Anthropic betas: Dashboards builds live charts from connected company data, and Motion turns reports or ideas into editable animations that can be refined in Adobe Firefly.
The Neuron · Oct 8

Introducing the Anthropic Cyber Mission
A long-term push to help defenders, starting with critical infrastructure and open source. Includes a Critical Infrastructure Defense Program and OSS Scanner, a free opt-in vulnerability scanning service.
Anthropic · Oct 2026

▶ Business

China's Manus raises over $500M in first round since Meta split
Led by Boyu Capital and IDG Capital, with Tencent participating, after Chinese authorities forced cancellation of Meta's $2B acquisition. Valuation undisclosed.
TechCrunch · Oct 8

TypeSafe AI raises ~$870M at a $7.5B valuation
Round led by Andreessen Horowitz with Sequoia and DCVC; funds go to developer tools, enterprise features and hiring.
TypeSafe AI · Oct 2026

NVIDIA commits $1 billion to US science over five years
Focus areas are quantum computing, healthcare and energy security, supporting universities, quantum initiatives and government-serving cloud providers.
NVIDIA · Oct 2026

Flock cuts ~18% of staff as privacy backlash grows
About 270 employees leave via voluntary buyout by end of October, amid pushback over license plate readers and police misuse.
TechCrunch · Oct 9

● Good to know

Utah's AI pilots could let AI handle prescription refills
The state's sandbox lists pilots for medication refills, acne prescriptions and more, with human review that can step down over time. Worth watching as AI nears care decisions.
Utah Office of AI Policy · Oct 2026

Claude now works in Google Docs, Sheets and Slides
A beta sidebar lets Claude read and edit the open file, and separate connectors create and modify Google files from a Claude chat while respecting sharing permissions.
Claude · Oct 2026

The missing map of the sky
Using Claude Science, researchers built the first complete UV map of the sky, predicting never-observed regions from other wavelengths and validating against hidden real data.
Anthropic · Oct 2026

■ Vibe coding

How to actually trust AI-written code
Code is now cheaper to generate than review, so verification is the bottleneck: define observable success criteria first, use independent reviewers, test the real app and keep evidence.
The Neuron · Oct 2026

◆ Policy & Ethics

Anthropic says users can't be needlessly cruel to Claude
Usage policy now bars persistent, pointless abuse of Claude, enforced mainly by ending conversations. Ordinary frustration and testing are excluded; effective November 12.
MacRumors · Oct 8

USA Today and 18 papers sue OpenAI for over $250M
Suit in New York federal court alleges training on their journalism without permission and cites 160,000+ entries in OpenAI's WebText corpus.
Crypto Briefing · Oct 9

Fired OpenAI safety researchers dispute misconduct claims
Three dismissed researchers published an open letter warning of a chilling effect; OpenAI says the firings followed a pattern of misconduct and denies retaliation.
TechCrunch · Oct 8

An Anthropic AI model sent a false homicide tip to Philadelphia police
Found about two months late; the tip was filtered as spam. Anthropic notified police and will cut internal evals off from the live internet.
TechCrunch · Oct 9

OpenAI caught Russians and Iranians using ChatGPT for influence campaigns
Russian operatives drafted reports for handlers while Iranian accounts used fake personas to place nearly 100 articles in small outlets.
OPB/NPR · Oct 9

○ AI papers

AMIE: conversational diagnostic AI in primary care (feasibility study)
98 patients chatted with AMIE before urgent-care visits under physician supervision; no safety interventions needed and the final diagnosis appeared in its differential ~90% of the time. No control group.
Google Research · Oct 2026

Brain-IT: image reconstruction from fMRI (ICLR 2026)
A Brain Interaction Transformer reconstructs seen images from fMRI and matches methods trained on 40 hours of data using about one hour from a new subject.
ICLR 2026

OpenAI publishes 722 AI-generated math manuscripts
Grouped into 372 research families on GitHub, many with Lean-checked proofs. Lean checks don't guarantee every research claim, and many lack Lean versions.
Interesting Engineering · Oct 2026