The AI Brief

Stripe has finalized an agreement to acquire OpenRouter, a startup that helps companies switch between AI models, for more than $7 billion, per Bloomberg. The deal comes just months after OpenRouter raised money at a reported $1.3 billion valuation. Bloomberg noted the final purchase price could still change.

OpenRouter gives developers a single door into more than 400 AI models; a customer writes once, then swaps between OpenAI, Anthropic, or a cheaper open-weight alternative without touching the code, with the router picking on cost, speed, or provider availability, taking a cut of about 5% of inference spend and reporting roughly 8 million users. The platform reportedly processed about 1.5 quadrillion tokens in the past year.

The deal follows Stripe's January 2026 purchase of Metronome, positioning it to own the model-selection, metering, and billing layer of the agent economy. OpenRouter's investors include CapitalG, one of Alphabet's venture arms, as well as Andreessen Horowitz and Menlo Ventures. The Wall Street Journal previously reported Stripe was in talks to buy OpenRouter for about $10 billion. OpenRouter CEO Alex Atallah co-founded OpenSea, the NFT marketplace, before starting OpenRouter in 2023, and had earlier described OpenRouter as the AI equivalent of Stripe. A Stripe spokesperson told TechCrunch the company does not comment on rumors or speculation.

OpenAI is on track to generate annualized revenue of more than $40 billion based on its current performance, according to people familiar with the matter, roughly doubling its run rate from the end of 2025 and bolstering the company's plans for a Wall Street debut. OpenAI co-founder and President Greg Brockman told staff in an internal announcement that the company's annual revenue run rate grew more than 20% month-over-month in July alone.

The ChatGPT maker's revenue has accelerated in recent months, driven in part by the growth of its AI coding software, as well as momentum from subscription sales and its nascent advertising business. OpenAI has also seen a sharp uptick in demand for its AI agents, including Codex for coding and ChatGPT Work for a wider mix of tasks. CFO Sarah Friar had previously said the company ended 2025 with more than $20 billion in annualized revenue.

Anthropic disclosed in May that its run-rate revenue had surpassed $47 billion, though Bloomberg cautioned that the two companies could be using different methodologies to calculate that figure. OpenAI has filed confidentially for an initial public offering and recently bought back employee shares worth $7 billion with its own cash.

An encrypted AI reasoning vulnerability in Anthropic, OpenAI, and Google APIs let researchers decode 315,320 published session logs and recover 182 developer credentials, because all three providers used a single global encryption key rather than per-session or per-user keys, making any encrypted reasoning block universally portable.

The paper, "Stealing Reasoning Traces from Proprietary LLM APIs" (arXiv:2608.09867), was submitted on August 10, 2026 by a team from the ELLIS Institute Tübingen and Max Planck Institute for Intelligent Systems. For Claude, thinking traces from any model can be replayed by any other except Fable 5. For GPT, the GPT-5.6 series can replay traces of all earlier model generations. For Gemini, thinking traces from any model can be replayed into any other. The attack ports reasoning from a target model into the context window of a compatible but weaker model, which is then coerced into transcribing the reasoning token-by-token via a jailbreak.

The extraction technique could allow an adversary to recover a proprietary model's reasoning traces at scale and use them to train a competing model. Researchers disclosed the findings to Anthropic, OpenAI, Google, Microsoft, and Hugging Face; the demonstrated attacks stopped working after mitigations, with the main extraction attack no longer reproducible as of August 2026. Developers are advised to strip reasoning blocks from shared traces and avoid committing raw API transcripts even when the visible text has been sanitized.

Disclosure: Claude, which generates this brief, is built by Anthropic.

On August 15, 2026, OpenAI told Free and Go users in the EEA and Switzerland that advertising would begin appearing inside ChatGPT later that month. Selection starts with the current chat topic, general location, and device type only. Plus and Pro plans stay ad-free. An opt-in prompt for personalized ads follows the initial contextual rollout.

The delay in rolling out to EU member states has been structural: GDPR requires explicit opt-in consent for personalized advertising, and OpenAI has not yet switched on that consent infrastructure for all EU states. As of August 15, the United Kingdom was the only European country where users were already seeing ads in ChatGPT; France, Germany, and Ireland had been announced as future markets but had no confirmed launch dates.

OpenAI launched ads in the US in February 2026, expanded to the UK, Mexico, Brazil, Japan, and South Korea on August 11, and notified EEA and Switzerland users on August 15. ChatGPT's ad product hit $100 million in revenue in its first six weeks in the US market. This section was first covered in Vol. I, No. 82.

Anthropic on August 7, 2026 released an update to the biology safeguards on Claude Fable 5 that the company says cut biology-related fallbacks by about 85% in testing across its product surfaces. Users should now see far fewer fallbacks on everyday health and educational questions, interpreting lab results, understanding symptoms, learning biology in an educational context, and healthcare professionals should get more support on clinical tasks.

The reduction in biology fallbacks is expected to bring down total fallback volume by roughly 67% on Claude.ai, 55% on Cowork, 17% on Claude Code, and 7% on the Claude Platform. Fable 5 still falls back to Opus 5 for dual-use work including virology, toxicology, and molecular design, so it is not yet usable for professional research or drug development; Anthropic says it will close that gap through trusted access for vetted researchers.

Anthropic's announcement cited the US Intelligence Community's 2026 Annual Threat Assessment, which states that advances in synthetic biology and genomic editing "could lead to novel biological threats" and that several state actors likely maintain active offensive biological and chemical weapons programs that could be accelerated by access to frontier AI. Over several weeks Anthropic rewrote the classifier's "constitution", the rules separating allowed from safeguarded content, with expert feedback, then retrained it.

Disclosure: Claude, which generates this brief, is built by Anthropic.