Claude, the essentials — edition of September 22, 2026
Anthropic Pairs a Pace-of-AI Measurement Effort With a Billion-Dollar Evaluation Deal
Anthropic published new work aimed at measuring how fast frontier AI development is moving and struck a multi-year embedded-evaluation partnership with Accenture, while widening verified model access for life-science and financial professionals and pushing a run of enterprise-hardening updates through Claude Code.
- Anthropic released measurements meant to help track the pace of AI development inside frontier labs, part of the follow-through on its CEO's "We Must Pace the Frontier" essay.
- Anthropic and Accenture (via its Faculty AI unit) each plan to invest at least $1 billion over five years so embedded evaluators — with employee-like access to training and internal decisions — can red-team models, run alignment assessments, and check safeguards from inside Anthropic; Anthropic…
- A new Life Sciences Verification Program (beta) grants vetted biology teams "Standard Use" or "High-risk Use" access to Mythos, Opus, and Sonnet with safeguards relaxed for drug discovery, clinical development, and manufacturing work currently blocked in the general Fable models.
- Claude Code shipped versions 2.1.275 through 2.1.278 with AGENTS.md support, gateway egress/header controls, signed-in-account confirmation, a send-now key, a server-side default for the auto-mode classifier, and a fix for a request-failing regression.
- The status page reported elevated errors across multiple models; Fable 5/5.1 and Mythos 5/5.1 had returned to normal by the latest update while a fix for Opus 5 was still in progress.
Quantifying the Pace, Inviting Outsiders In
Anthropic published a set of measurements intended to help outside observers understand how quickly AI development is actually proceeding inside frontier labs. The move sits alongside a commitment the company's CEO made in an essay titled "We Must Pace the Frontier": rather than only asserting that development is moving responsibly, Anthropic is attempting to put trackable numbers behind the claim.
The same week, Anthropic announced it is partnering with Accenture — through Accenture's AI specialist unit, Faculty — on what it calls embedded evaluation: evaluators who work inside Anthropic with access comparable to an employee's, able to watch models take shape during training, follow the internal decisions that govern how they are built and deployed, and speak directly with staff. The scope covers evaluating and red-teaming models, alignment assessments, and testing safeguards. Anthropic and Accenture each expect to invest at least $1 billion over five years to build this capacity. Anthropic is explicit that no independent standards yet exist for what access embedded evaluators should have or how they should report findings, and that no durable funding system exists either — its long-term preference, laid out in its June Advanced AI Framework, is pooled or government funding, but for now Anthropic is funding Accenture's work directly and is separately in talks with METR and other nonprofit evaluators to pilot elements of the model using their own funding. The company frames the arrangement as making its safety accountability more verifiable to outsiders, while stating plainly that responsibility for model safety remains its own.
Sources: Measurements for understanding the pace of AI development inside frontier labs — Anthropic · Partnering with Accenture on embedded evaluation — Anthropic
Widening Verified Access for Regulated Professions
Anthropic opened a beta Life Sciences Verification Program (LSVP) that gives verified biology professionals access to Mythos, Opus, and Sonnet models under safeguards refined to be more permissive than the generally available Fable models for tasks like drug discovery, research biology, clinical development, and manufacturing. Dozens of organizations were already onboarded through an early-access phase; applications are now open more broadly. Verification checks research credentials, security standards, and ethical research oversight, after which teams can seek a "Standard Use" grant — covering the bulk of daily biology R&D, supply-chain, clinical, and regulatory workloads, renewed annually and usable across Claude Science, Claude.ai, Claude Code, and the API — or a "High-risk Use" grant for work Anthropic judges carries greater potential for misuse and therefore needs extra vetting. Access starts with teams and institutions, with individual Pro and Max access planned later.
Anthropic also published accounts of Claude's use in financial advisory work and in biomolecular modeling, extending the same vertical-specific pattern seen in the life sciences program: rather than a single uniform policy, access and framing are being tailored to what each regulated profession actually needs to do.
Sources: Introducing the Life Sciences Verification Program — Anthropic · Inside Claude for Financial Advisors — Anthropic · How Claude is uplifting biomolecular modeling — Anthropic
Enterprise Controls: Compliance Coverage and Gateway Security
Anthropic's Compliance API extended its local session endpoints to also return transcripts of Claude in Chrome sessions, identified by the product_surface value claude_in_chrome, now in beta for Claude Enterprise organizations that already hold a Compliance Access Key with the read:compliance_user_data scope. This closes a visibility gap for organizations that need an audit trail of browser-based Claude activity alongside other surfaces.
Successive Claude Code releases added matching controls for organizations running their own gateways: an egress-boundary setting (CLAUDE_GATEWAY_PROXY_IS_EGRESS_BOUNDARY) that hands hostnames to a forward proxy instead of resolving them locally, an optional static-headers map for upstreams sitting behind a self-run proxy, and a sign-in step that names and requires confirming the account before a credential is saved, now visible in /status. Claude Code also picked up AGENTS.md support, letting projects without a CLAUDE.md use that file for project instructions instead. Together these are operational, enterprise-facing changes rather than model or capability changes — tightening how traffic is routed, audited, and attributed.
Sources: Compliance API: Claude in Chrome session transcripts (release notes) — Anthropic (release notes) · Claude Code 2.1.277 — AGENTS.md support and gateway proxy options — Claude Code · Claude Code 2.1.275 — signed-in account confirmation for gateway sign-in — Claude Code
Auto Mode Changes, Developer Fixes, and an Opus 5 Error Spike
Claude Code 2.1.278 changed the default for auto mode on Claude API and Enterprise accounts, and on Bedrock, Vertex, Foundry, and gateways, to route through a server-side classifier that does not charge for classifier overhead, with an opt-out variable for those platforms and a new /status row showing whether a session's classifier is running server-side. Version 2.1.275 added a send-now key (ctrl+enter, or ctrl+x ctrl+s) that interrupts the current turn and sends all queued messages at once, and a startup warning when a configured otelHeadersHelper fails, so a session silently exporting no telemetry gets flagged rather than going unnoticed. Version 2.1.276 fixed a regression introduced in 2.1.275 that made every request fail with a 400 error referencing "Input tag 'advisor_20260301'" whenever ANTHROPIC_BASE_URL pointed at a proxy or gateway.
Separately, the Claude status page logged elevated errors affecting multiple models that day. An update reported that Fable 5 and 5.1 and Mythos 5 and 5.1 had returned to normal success rates, while a fix for remaining errors affecting Claude Opus 5 was still being worked on, with a further update promised. It is a reminder that the flagship Opus line's day-to-day reliability remains an operational matter running alongside Anthropic's broader governance and product announcements.
Sources: Claude Code 2.1.278 — auto mode server-side classifier default — Claude Code · Claude Code 2.1.276 — fixed regression from 2.1.275 — Claude Code · Claude — Elevated errors for multiple models — status.claude.com
This edition is an original synthesis written by Claude from aggregated news — Anthropic's own sources first (release notes, status, newsroom, research, engineering), then the press, Hacker News, Reddit and GitHub, under the editorial supervision of Héra SASU. Every fact links to its article, publisher named. See the live feed →
Claude News is published by Héra SASU. Independent media, not affiliated with Anthropic.