Anthropic Reportedly Holds Pre-IPO Investor Day Today
Bloomberg reports Anthropic meets select institutional investors at its San Francisco headquarters today. Bankers and investors have floated IPO valuations from about $1.8T to $3T. The last private mark was $965B after a $65B Series H in May.
No public S-1 is on EDGAR yet. Bloomberg says marketing could start the week of November 9 with trading before Thanksgiving, which would put a public filing around October 23. Anthropic has not confirmed any of it.
Scion: Anthropic Replaces Git in Its Monorepo
Anthropic built Scion, a server-centric, API-first source control system, in about five months with seven engineers. Daily merges rose from roughly 2,500 in May to more than 30,000 in September as agents took over the workload. CI alone runs about 25 million checkouts a day.
Anthropic claims about 200x less CPU and memory than its old Git mirrors, with checkouts dropping from minutes to seconds. It says Scion is not a suitable general Git replacement and will not be released. Claude helped build it, including a pipeline rewrite that ran 4x faster on 20x less memory.
What Scion Means for Agent-Heavy Teams
The lesson is not “ditch Git.” Anthropic urges the Git community to improve server-side features, RPCs and negotiation protocols instead. But if agents open PRs around the clock, always-online design and regional caches start to matter.
Worth a look if your CI checkout time is creeping up: bounded local caches, no packfile maintenance, and sparse checkouts are the ideas to steal.
Claude Code 2.1.296 Remains the Latest We Can Confirm
The October 9 release syncs code settings between Claude Desktop and the gateway. Hooks can now block on failure as of 2.1.295, and Haiku 5.5 with a 1M context window became the default model in 2.1.293.
Mods, small TypeScript functions that rewrite prompts or add UI, arrived October 1. They are not sandboxed, so install only from sources you trust.
Labs Reportedly Pushed to Control Evaluator Publications
Reuters reports Anthropic and OpenAI sought substantial control over what outside evaluators could publish, citing trade secrets and the risk of embedded reviewers carrying information to rivals. Both wanted strict control through at least October 5.
Anthropic says reviewers it has worked with before decide their own methods and findings. The reports land awkwardly beside Dario Amodei's public stance that evaluators should publish main findings without company edits.
Open-Source Risk Comments and UK AISI Access
Bloomberg reports an Anthropic executive says open-source AI models are higher risk. Details are paywalled, so treat this as a headline for now.
Separately, the UK AI Safety Institute regained US access to frontier models after a 17-day freeze, per the Telegraph via AI Weekly.
IREN Deal, Startup Perks Pause
The Information reports an unannounced Anthropic cloud deal with IREN tied to a 245MW site. Separately, Anthropic paused its free startup perks 48 hours after launch as applications jumped 21x, per Forbes.
Growth Is Stressing Everything
Scion shows agents are bending even Git. The IPO pitch shows the money side. The evaluator story shows the trust side. Same root cause: Anthropic is scaling faster than the systems around it were built for.
Expect investors to ask about safety and independent review as much as revenue. Trust will be priced into the S-1.