What is Claude Mythos and why was it kept secret?: Claude Mythos Preview is Anthropic's most capable model, released 8 April 2026 exclusively to 17 Glasswing partners. Anthropic withheld it from public release because it can chain 32-step cyberattacks equivalent to 20 human work-hours, raising systemic risk concerns. The UK's AISI independently confirmed the capability.Source: Lowdown
What is Project Glasswing?: Project Glasswing is Anthropic's restricted access programme for Claude Mythos, expanded to 17 partner organisations in April 2026, each receiving $100M in model usage credits. Named partners include JPMorgan Chase, Broadcom, CrowdStrike, Nvidia, Palo Alto Networks, and Cisco.Source: Lowdown
What is Anthropic?: Anthropic is a US AI safety company founded in 2021 by Dario and Daniela Amodei and other former OpenAI researchers. It develops the Claude family of models, with Amazon as its largest backer at up to $4 billion.Source: Lowdown
Why did Tom's Hardware challenge Anthropic's Mythos claims?: Tom's Hardware found that Anthropic's claim of 'thousands of zero-days' rested on only 198 manual reviews of largely outdated software. The UK's AISI later confirmed chaining capability (32 steps, 20 hours) while also finding no single-task superiority, partially vindicating the critique.Source: Tom's Hardware / AISI
What is Claude Cowork and why did it crash India's IT stocks?: Claude Cowork, launched January 2026, is an autonomous AI agent built to handle workplace tasks end-to-end. Its launch wiped $24 billion from India's Nifty IT index in a single session, exposing the sector's vulnerability to autonomous AI agents.Source: Lowdown
What is Anthropic's deal with News Corp worth?: News Corp CEO Robert Thomson disclosed on 8 May 2026 that the company anticipates a $1.5 billion settlement with Anthropic. The figure was described as anticipated and not yet formally closed.Source: Lowdown briefing 2026-05-10 / News Corp Q3 FY2026 earnings call
Why does Anthropic's scrape ratio matter for publishers?: The Center for Journalism & Liberty found Anthropic's scrape ratio at 73,000 pages returned per human visitor, compared to OpenAI's 1,700:1. The higher ratio means Anthropic extracted proportionally more journalistic content relative to real users, strengthening publishers' claims for larger compensation.Source: Center for Journalism & Liberty, April 2026
How does Anthropic compare to OpenAI in publisher licensing disputes?: Both companies face publisher licensing claims, but Anthropic's disclosed settlement figure of $1.5bn (anticipated, News Corp, May 2026) significantly exceeds OpenAI's previously reported deals. Anthropic's scrape ratio (73,000:1) is FAR higher than OpenAI's (1,700:1), according to the CJL April 2026 report.Source: Lowdown briefing 2026-05-10 / CJL April 2026

Background

In the publisher-licensing market, Anthropic is the subject of what is currently the largest single publicly named AI content deal: News Corp chief executive Robert Thomson disclosed on 8 May 2026 that News Corp anticipates a $1.5 billion settlement with Anthropic — an 'anticipated' figure read into an SEC-filed earnings transcript, not yet a closed agreement. The Center for Journalism & Liberty's April 2026 report identified Anthropic's scrape ratio at 73,000 pages per human visitor returned, versus 1,700:1 for OpenAI — a ratio that shapes Anthropic's relative exposure in publisher litigation and explains why its settlement figure exceeds News Corp's previously disclosed five-year OpenAI deal by roughly six times.

Source Material

Anthropic coverage — Reuters Anthropic — Wikipedia Anthropic Official Website

How the World Sees Them

India

India's IT sector was the first major casualty of Anthropic's product ambitions. The Claude Cowork launch wiped $24 billion from Indian IT stocks in a single session, exposing the sector's vulnerability to autonomous AI agents.

Tech industry

Competitors OpenAI and Microsoft are racing to match Anthropic's agentic AI offerings. Anthropic's safety-first positioning and Amazon backing make it a formidable rival despite being newer than OpenAI.

United States

Anthropic is headquartered in San Francisco and founded by former OpenAI researchers, placing it at the centre of US AI policy debates. Its CEO's call for a robot tax aligned it briefly with Senator Sanders, an unusual coalition for a Silicon Valley firm.

Labour economists

Anthropic's own research introduced the concept of "observed exposure" to measure real AI displacement. Findings showing 75% task coverage for programmers gave empirical weight to displacement fears previously dismissed as speculative.