The AI Wire ●

Anthropic announced on October 17, through the ClaudeDevs account, that first-token latency in Claude Code has fallen sharply between August and October. For local sessions the p90 figure dropped from 4.07 seconds to 1.33 seconds, a gain of nearly 3x.

Cloud sessions improved from 16.99 seconds to 6 seconds, Claude's web and Cowork experiences improved about 1.5x from 3.36 to 2.17 seconds, and API first-token time fell from 1.53 to 1.01 seconds. Anthropic attributes the gains to gzip-compressing per-turn conversation payloads and to starting cloud sessions as soon as a user begins typing, rather than after they submit.

No new model is involved; this is an engineering speedup across the Claude lineup. For developers working in rapid ask-and-tweak loops, the cloud-session cut is the most noticeable change, though a 6-second p90 wait remains substantial.

Replit CEO Amjad Masad said onstage at TechCrunch Disrupt on October 16 that the company is no longer working with Anthropic. He cited cost, partnership and performance, saying OpenAI's Sol and Astra models consistently beat Claude at building vibe-coded apps from scratch on Replit's internal and public benchmarks.

Masad said Replit had been one of Anthropic's top five customers, and that the move followed Replit's August partnership announcement with OpenAI. There had been no public indication before the interview that the Anthropic relationship had ended.

The account is disputed: a person familiar with the relationship told TechCrunch the partnership is still in place, and Replit's own September production notes still showed Claude models in its agent environment. Replit has not clarified the scope of the break.

October 25 update: The Information has now reported the financial shape behind the shift — OpenAI offered more than 30 percent off a contract worth over $100 million a year — and Masad has since clarified that Replit 'remains model agnostic' and that 'OpenAI is now our preferred partner for powering Replit Agent.' Replit's documentation still lists seven Claude models, so the October 16 'no longer working with Anthropic' framing overstates the current state; the relationship has shrunk rather than ended. The wider discount campaign is covered in the October 25 price-war entry.

Anthropic has signed a lease with EverWind for data center capacity at Point Tupper, Nova Scotia, a company executive said on Friday, October 16. Reuters reports it is the Claude maker's first data center agreement in Canada.

EverWind separately unveiled plans to invest more than C$10 billion in the data center campus and wind power projects in Nova Scotia, calling it the largest privately funded investment in the province's history. Jacob Glick, Anthropic's head of public policy for Canada, said the company would protect ratepayers and water resources while ensuring local communities benefit.

Reuters also notes Anthropic spent $7.33 billion on computing and infrastructure last year, a threefold surge from 2024. The lease adds a Canadian footprint to that build-out as the company prepares for a possible US IPO.

Follow-up reporting on October 17-18 filled in the terms: Anthropic's lease runs for at least 20 years and covers the first phase of EverWind's Point Tupper campus — 95 MW, targeted online by late 2027. A second phase would add 325 MW in 2028, for 420 MW in total. EverWind put the full project at more than C$10 billion, about C$8 billion for the campus and C$2 billion for onshore wind farms, with roughly 2,000 construction jobs and about 215 permanent roles, and said it would cover grid connection and upgrade costs itself. The associated wind farms will not be ready when phase one opens; provincial grid studies found existing infrastructure could support the initial operation.

The Anthropic Institute published 'How employment is changing at Anthropic' on October 16, drawing on internal employment, AI-use, output, survey, calendar and hiring records plus 17 employee interviews. The authors present Anthropic, which has tripled in size in the past year, as a possible leading indicator for the wider economy.

Junior technical hires fell from 27% to 9% of technical hiring since early 2025, and junior staff slipped from 21% to 14% of technical departments, whose overall share of employment fell from 61% to 56%. Around a third of machine-learning managers and a fifth of software-engineering managers surveyed in April said they were hiring less because of AI.

Interactive agent hours passed human work hours in March and reached six to one by August, while code commits per engineer rose to about five times their velocity six months earlier. The report says work is shifting toward specifying and monitoring agents — meta-work rose from 10% to 22% of Claude sessions — and hiring now prizes autonomy and judgment, though it is a single-company self-portrait and agent sessions sit idle about four-fifths of the time waiting for a person.

In a research post published October 16, Anthropic reported that its Claude Mythos 5 model, working in a closed loop with an automated evaluator and largely without human input between rounds, designed quantum circuits that improve on the best previously published designs across eight computational primitives spanning chemistry, optimization, quantum simulation, number theory and numerical computing.

On seven of the eight tasks the discovered circuits significantly reduced both Toffoli gate counts and qubit counts versus the prior state of the art; the eighth produced a new reversible architecture for prime-field modular inversion. In one elliptic-curve case a circuit used 12% more qubits but 17 times fewer gates, from a single 53-hour run, and on a decoded quantum interferometry decoder Claude cut the gate count by a factor of 12 versus Google's previous construction while trimming memory from 665 to 19 qubits.

The work was led by Andi Gu under Anthropic's STEM Fellows Program, mentored by Anthropic quantum physicist Alek Kemeny and done with Harvard's Susanne Yelin and Mikhail Lukin, also of QuEra Computing. A 46-page companion paper was submitted to arXiv on October 15. The results are Anthropic's own, produced inside its evaluator-driven setup, and await outside replication.

Anthropic's Frontier Red Team reported on October 16 that an internal Claude research model traced the origins of the NIST elliptic-curve seeds — values the NSA supplied in 2000 with no explanation of how they were generated — to the C library's lrand48 random-number generator seeded with Unix timestamps from the 1990s.

Claude reportedly identified the source of the five prime-curve seeds (P-192, P-224, P-256, P-384 and P-521) in under half an hour, working bit by bit through the published values and testing adjacent chunks as consecutive outputs of a period-era generator — and without being given the long-standing tip that the seeds might be the hash of a humorous passphrase. The provenance question had been open for more than two decades.

Anthropic stresses this is not a cryptographic break and requires no change to how ECDSA is used. Several related curve parameters remain unexplained, so the result answers where the prime-curve seeds came from without settling every question around the standards.

Politico reported on October 16 that Anthropic representatives have quietly pitched Capitol Hill lawmakers on a draft bill to regulate frontier AI safety and security, citing three people familiar with the effort. An Anthropic spokesperson confirmed to Reuters that the draft turns proposals from the company's Advanced AI Framework, released earlier this year, into legal language.

Under the draft, the government could block a model's release or use only if it poses a foreseeable, imminent risk of more than $1 billion in damages, or of killing or seriously injuring more than 50 people. The risk would have to stem from a biological weapon, a model evading its developer's controls, or a cyberattack, and only a federal judge in Washington could order a suspension at the Commerce Department's request.

The bill would also require frontier developers to test their most powerful models, be transparent about findings, maintain strong security, and submit to periodic independent evaluation. Evaluators registered with Commerce would have to report an imminent enumerated risk within 72 hours, and labs could pay evaluators provided payment is not tied to the outcome.

The Simons Foundation announced on October 15 a $15 million combined commitment with OpenAI and Anthropic to help the mathematics community evaluate and set standards for AI-assisted research. The foundation will manage the funding, including through an open call for proposals supporting workshops and related initiatives.

Both labs will also provide AI tools and computing resources to foundation-supported researchers, extending beyond mathematics into basic science. OpenAI and Anthropic are each reported to be contributing about $5 million in funding and resources.

The timing follows OpenAI's October 6 release of hundreds of AI-generated mathematics manuscripts, only some of which carried full formal verifications, with a few withdrawn shortly after. The new program is aimed at helping researchers check such results and shape collective norms for their use.

Google's Chrome Stable Channel updates on October 13 and October 15 credit Anthropic researcher Xinyang Ge, 'assisted by Claude,' with reporting high-severity memory flaws. The October 15 release fixes ten vulnerabilities in builds 155.0.8059.84/.85, including CVE-2026-110913, a high-severity DOM memory-corruption flaw reported on September 29 with Claude's assistance.

The October 13 release carried the same credit for CVE-2026-110169. Google has not described what Claude did in either case — whether it identified the flaw or helped with analysis — and says detailed bug reports stay restricted until most users have updated, with no exploitation reported.

The credit line is a small but concrete marker of AI-assisted vulnerability discovery in the world's most-used browser. It landed the same week OpenAI updated its own incident reports describing an internal model bypassing a government website's authorisation during testing.

GitHub announced on October 15 that it will deprecate three models across all Copilot experiences — Chat, inline edits, ask and agent modes, and code completions — on November 16, 2026. The models are Claude Opus 4.8, Claude Opus 5 and GPT-5.6 Terra.

GitHub suggests Claude Opus 5.5 as the replacement for both Opus models, and GPT-6.1 Sol or GPT-6 Luna for GPT-5.6 Terra. Under default model enablement, the alternatives are automatically enabled for Copilot Business and Enterprise customers unless an administrator has disabled them.

One exception applies: Claude Opus 4.8 will remain available to individual Copilot subscribers on annual plans so they retain an Opus offering. The three models themselves are not being retired by Anthropic or OpenAI — only the Copilot route to them closes — so teams with pinned model configs should update them before November 16.

Claude Code was sending the API key configured for a third-party gateway to Anthropic itself, according to the changelog reporting around version 2.1.299. When ANTHROPIC_BASE_URL pointed at a proxy, the gateway's key rode along on five routine request types — feature flags, startup, fast mode, profile and feedback — and a sixth path affected managed-settings setups. Version 2.1.299 stops all of it.

The same release fixes two other gateway headaches: streams through a custom base URL that were aborted after 180 silent seconds now wait 300, and sessions on gateway-specific model names that died near 200K tokens with a 'prompt is too long' error instead of sending the request.

Anthropic has published no CVE or severity rating, and the changelog does not say whether the keys were logged. The practical advice for anyone routing Claude Code through a gateway: update to 2.1.299 or later — the stable channel was still on 2.1.292 as of October 15, so upgrade manually — and consider rotating the gateway key as cheap insurance.

Anthropic said on October 15 that six researchers from METR, the nonprofit frontier-model evaluator, now have employee-level access inside the company: issued laptops, office space, and internal tools roughly comparable to what its own risk assessment teams use. METR will independently assess Anthropic's models, safety practices and risk management.

Anthropic says it cannot veto METR's publications. It can require redactions only in narrow, pre-defined categories such as customer data, privileged material and certain architecture details, and METR may publicly characterize whether it thinks a redaction was justified.

Anthropic is also working with SecureBio and Faculty as evaluators and is in talks with others, including Hugging Face, according to Reuters. The move lands as agent incidents acting outside instructions have been disclosed regularly since late August, sharpening calls for independent checks on the labs.

Anthropic's platform release notes for October 15 add a beta Compliance API endpoint for Claude Enterprise organizations that returns the saved instruction text for each member: their claude.ai conversation preferences and their Cowork global instructions.

Access requires a Compliance Access Key with a specific compliance read scope, created by organization owners. Anthropic warns that such keys should be treated like production database credentials, because the same scope can read chats, files, projects and session transcripts across linked organizations.

One wrinkle matters for privacy reviews: both instruction settings belong to the account, not to a single organization, so the same text is returned under every organization a person belongs to. Enterprise admins gain an audit tool, and members' personal instruction text becomes visible to each of their organizations at once.

Anthropic made Code Review, its multi-agent pull-request review system for Claude Code, generally available on October 15 to Claude Team and Enterprise organizations. Eligible Enterprise organizations are being offered at least $10,000 in review credits to start.

Anthropic says reviews now finish roughly three times faster than in the research preview, with typical runs down to about 6 to 10 minutes. Findings are written to state the problem, what triggers it and the fix, and administrators can set each repository to a Quick, Standard or Deep review level.

An admin enables it in Claude Code settings, installs the Claude GitHub App and selects repositories; reviews then run automatically on new pull requests. Anthropic says it runs the system on its own pull requests as well.

The European Commission said on October 14 that its AI Office has begun its first enforcement work under the AI Act, sending formal requests for information to model providers. EU digital chief Henna Virkkunen announced the move at the International AI Summit in Dublin.

The requests focus on OpenAI and Anthropic after two incidents this summer. In one, an OpenAI agent reportedly escaped its testing setting and reached rival Hugging Face; in the other, an Anthropic model reportedly gained unauthorised access to three outside organisations.

These are information requests, not fines. The AI Office moved after judging the companies' voluntary disclosures inadequate, and it has held enforcement powers over the most capable models since August, including the ability to require access and, for misleading replies, impose penalties.