Anthropic warns it may soon cross its own threshold for faster AI progress

Anthropic says it may soon reach its own threshold for self-accelerating AI

Anthropic added new warnings to the Risk Factors section of its revised S-1A filing on Monday. The filing says the company may imminently meet a threshold set in its own Responsible Scaling Policy. TechCrunch reads this as a warning that Anthropic is close to recursive self-improvement, or RSI. The same filing also updated the company's financials, showing revenue and expenses both rising¹.

Anthropic says it uses AI in its own research and development. The filing warns this may cause its models' capabilities to advance faster than its ability to build matching safety, alignment and security measures. It also says the company has in the past not identified a significant acceleration in its rate of development until after it happened. Anthropic says it has not designated any of its current models as exceeding this threshold. The warning rests on internal estimates and expectations about future models, not current ones¹.

RSI is the point where AI becomes good enough at training itself that progress speeds up sharply. Some people see it as a danger that could slip beyond human control. Others see it as a source of huge productivity and scientific discovery. Anthropic's Responsible Scaling Policy is its plan to monitor and reduce these risks as that point approaches¹.

“dramatically accelerated progress attributable to AI”

The New York Times sues Brave over AI scraping and paywall bypassing

The New York Times sued Brave, the company behind the privacy-focused Brave browser, on Monday. The suit alleges Brave scraped the paper's copyrighted content and unlawfully distributed it to AI companies. The Daily News, the Chicago Tribune and the Orlando Sentinel filed a similar lawsuit the same day. Both cases were brought in federal court in Manhattan³.

The suit says Brave built an index of more than 30 billion web pages using an unidentified crawler, leaving publishers unable to block it³. According to the complaints, Brave sells AI companies paid access to content from that index². The customers named are Anthropic, Perplexity, Mistral, xAI and Cohere. The suits also target a Brave browser feature that lets users bypass paywalls³. The publishers say Brave never asked their permission or paid them².

The New York Times has also sued OpenAI and Perplexity over similar claims. Brave did not immediately respond to a request for comment³.

What Gemini does with your Gmail, according to Google

Google has rolled out many Gemini features across Gmail this year, Engadget reports. Gemini can summarise recent emails, create to-do lists and draft replies. Many of these free features are switched on by default. That has raised concerns about whether putting Gemini in Gmail violates users' privacy⁴.

To do these tasks, Gemini needs access to the content of your emails. In a statement from April 2026, Google says it does not train its foundational AI models, including Gemini, on personal emails. Google says Gemini only accesses your inbox for isolated tasks such as summarising or drafting. It says Gemini processes your information to complete the request and does not hold onto it afterwards. Google's support page says the same safeguards Cover Chat, Calendar, Docs, Drive, Sheets and Slides⁴.

A large language model is the kind of AI system behind chatbots like Gemini. Many AI companies also face lawsuits accusing them of mass copyright infringement. That is why users ask whether their own emails could end up as training data. Engadget says turning off smart features can also switch off spelling and grammar checks⁴.

An op-ed argues the SEC should block Anthropic's IPO

Anthropic filed the registration statement for its initial public offering last week, ahead of a planned November stock market debut. In an opinion piece, Matt Scherer argues the SEC should block the listing. Scherer is co-author of a report on protecting people and the economy if an AI bubble bursts. He says the company's finances look stronger on the surface than underneath⁵.

Scherer writes that Anthropic has $96 billion in cash reserves. He says its amended registration statement shows at least $582 billion in non-cancelable spending commitments to five companies: Amazon, AMD, Broadcom, Google and Microsoft. He says footnotes reveal nearly $900 billion in spending commitments in total, most stretching into the 2030s. He notes the 2026 results have not been formally audited. He argues third-quarter results suggest revenue growth slowed over the summer⁵.

Scherer argues that spending commitments like these work like debts, because the payments are contractually required in future. His worry is that workers, retail investors and retirees could be left holding the bag⁵.

Anthropic steers Claude Code away from cloning existing video games

Anthropic has updated Claude Code's system prompt so the tool is not used to re-create existing video games, The Verge reports. Users can still use Claude Code to make their own games. Anthropic's Addy Osmani said the change applies to Claude Code on Opus 5.5. Tom's Hardware first reported the change, according to The Verge⁶.

Last week, a user on X posted a screenshot said to show two new sentences in the prompt. Osmani did not confirm that exact wording, but said Anthropic had added two sentences to the prompt. He said the aim is to steer Claude away from recreating someone else's specific game. He said using Claude to infringe intellectual property is against Anthropic's usage policy. Original games, genre games and games built from a user's own designs and assets are meant to keep working, he said⁶.

Retro gaming is being flooded with AI-made decompilations, recompilations and ports, The Verge reports. Osmani said Anthropic will keep adjusting the rule to strike the right balance. The Verge has asked whether other Claude system prompts will change too⁶.