Anthropic courts investors ahead of a blockbuster fall IPO
Anthropic owns the news cycle this morning: the lab is on a pre-IPO roadshow that could produce the biggest listing in history, and it just switched on invisible watermarks for every new Claude model — transparency work aimed squarely at the scrutiny a public debut will bring.
Anthropic is meeting with potential investors to shore up confidence ahead of a planned IPO that could launch in September or early October — what the Wall Street Journal describes as potentially the biggest IPO ever. The company is touting rapid growth and laying out plans to address mounting public backlash against AI, according to people familiar with the matter, but the roadshow has also fielded tough questions about cheaper Chinese rivals, US-China tensions, and whether the AI infrastructure boom can sustain the revenue assumptions behind a near-trillion-dollar valuation. Anthropic's last private round priced it at $965 billion — comfortably above the $852 billion mark OpenAI just used for its $7 billion employee tender offer (we covered that this morning — OpenAI closes $7B tender offer at $852B valuation). The questions, not the answers, are the story: the first AI mega-IPO will price on whether institutions believe frontier-lab growth outlives the compute buildout that funds it, and the Journal's report suggests a meaningful slice of them are still skeptical. If the fall timeline holds, Anthropic would beat OpenAI to the public markets — a race that got sharper this week as OpenAI's own valuation went flat and its ethics chief walked out the door.
Anthropic has begun embedding invisible watermarks in text generated by new Claude models — the most sweeping text-watermarking rollout by a frontier lab, driven by the EU AI Act transparency rules it signed under the bloc's Code of Practice. Models launched in the EU on or after August 2 weave an imperceptible, machine-readable signal directly into the text itself rather than into metadata, so it survives copy-paste and "may persist through some editing"; supported files such as images carry digitally signed provenance metadata as well. The markings apply everywhere Claude is offered — the API, chat, Claude Code, Claude Cowork — and Anthropic says existing models will gain them later. The reception shows why this is a test case: critics on X dismissed it as a fingerprint anyone can rephrase around, and Paul Graham floated a startup idea for stripping the marks by rewriting output — the same cat-and-mouse dynamic that has dogged every previous watermarking attempt. For readers, the practical shift is real but modest: provenance signals only prove Claude touched the text, not that a human didn't write it (we covered the EU transparency rules coming into force in August — EU AI Act transparency rules now enforceable in Europe).
What to watch: whether the roadshow's tough questions force Anthropic to concede on valuation or float size — and whether OpenAI files its S-1 publicly before Anthropic prices.
Should every frontier model's output carry an invisible watermark — and would rephrasing to strip it be fair game? Tell us in the comments.
Sources: The Wall Street Journal · Techmeme · Newsquawk · Anthropic · QbitAI · Interesting Engineering · India Today
