Anthropic News | White House Opposes Mythos Expansion Plans

[1] White House Officially Opposes Anthropic's Mythos Expansion Plans

On April 30, it became public that Trump administration officials told Anthropic they could not agree to plans to expand access to "Claude Mythos Preview" — an unreleased model for defensive cybersecurity — to approximately 70 companies and organizations. Mythos is said to be capable of penetration testing, vulnerability discovery, and malware reverse engineering, and the government expressed concern about "the risk of misuse if overly powerful offensive capabilities were to leak to external parties." This is regarded as an early case of export-control-adjacent debate surfacing in a commercial SaaS context.

Government Concerns and Industry Implications

What government officials are objecting to is not Mythos's capabilities themselves, but the breadth of who would have access. Reports indicate the 70 companies include not only defense contractors but also private security vendors and some critical infrastructure operators, and concern was raised about commercial distribution proceeding before an authentication framework is in place. On the same day, a competing major AI company announced it would roll out a rival cyber-specialized tool, "GPT-5.5 Cyber," exclusively to "critical cyber defenders." Given that this company had previously criticized Anthropic's restrictive approach, it is now clear that the industry as a whole is moving to align on deployment lines for cyber AI. Organizations that had planned on commercial deployment of Mythos may be forced to revisit their distribution lists, terms of use, and audit log requirements, and adoption plans for SOC or Red Team use are likely to enter a short-term holding pattern.

Impact: ★★★

Source: TechCrunch 2026-04-30

[2] 1M Token Context Window Beta Officially Discontinued (Affects Sonnet 4.5 / Sonnet 4)

As of April 30, 2026, the 1M token context window beta (beta header: context-1m-2025-08-07) for Claude Sonnet 4.5 and Claude Sonnet 4 has been officially discontinued. Anthropic announced the planned discontinuation as of March 30, with April 30 as the implementation date. Going forward, attaching the beta header to these models will have no effect, and requests exceeding 200k tokens will return an invalid_request_error — meaning behavior in production environments changes immediately.

Confirming Migration Targets and Pricing

To continue using 1M tokens, migration to Claude Sonnet 4.6 or Claude Opus 4.6 is required. Both models offer 1M tokens in general availability with no additional headers required and at standard pricing, with no special beta program enrollment needed. For workloads that regularly consume more than 200k tokens — such as long-form RAG, full codebase ingestion, or large-scale log analysis — it is safest to simultaneously swap the model identifier (e.g., claude-sonnet-4-5 → claude-sonnet-4-6) and recalculate cost differences based on per-token pricing.

Migration Checklist

The three things to confirm on the API side first are: (1) whether you can remove anthropic-beta: context-1m-2025-08-07 from request headers, (2) whether any batch jobs returning inputs larger than 200k tokens remain, and (3) recalculating expected response lengths and costs. Depending on your SDK version, model constants may be outdated, so update @anthropic-ai/sdk and the anthropic Python SDK to the latest versions and monitor error logs for token rejections over 200000 for about a week.

Impact: ★★★

Source: Anthropic Platform Release Notes

[3] Reports of Anthropic Raising ~$50 Billion at ~$90 Billion Valuation

According to TechCrunch reporting dated April 29, multiple investors have reportedly approached Anthropic about raising approximately $50 billion at a valuation of $85–90 billion. This represents a significant increase from the $38 billion valuation in February, attributed to annual revenue approaching $40 billion. A more-than-doubling of valuation in under three months is extremely rare even in the generative AI space and signals an accelerating influx of capital into the industry.

Context of the Funding Round

Management is reportedly set to decide on a course of action at the May board meeting, and it has been reported this could be the last private round before a public offering. Investor demand is described as extremely strong, with even institutions prepared to invest $5 billion reportedly unable to secure a meeting with the CFO. The primary intended uses of the raised funds are scaling up proprietary inference infrastructure, extending data center contracts, and expanding model research headcount — making this a raise that is largely defensive in nature against rivals in the race for compute resources.

Implications for Developers and Enterprise Users

If a large-scale fundraise is completed, concrete improvements such as greater API stability, expanded rate limit quotas, and shorter new model release cycles can be expected. On the other hand, the sharp rise in valuation could also affect future API pricing structures and enterprise contract minimum commitments, so it is prudent for companies considering long-term contracts to watch for any pricing updates after the fundraise is announced. The Clauder Navi editorial team plans to publish a dedicated follow-up article once the fundraise is officially announced.

Impact: ★★

Source: TechCrunch 2026-04-29

[4] Claude Code v2.1.125–v2.1.126 Released — New Commands and Multiple Bug Fixes

In v2.1.125, released April 29, a bug was fixed in which the allowManagedDomainsOnly / allowManagedReadPathsOnly security settings were ignored under certain conditions. These flags are critical settings that control whitelisting of managed domains and readable paths; when disabled, unintended communications to outside domains or access to unexpected paths could occur. If you rely on these settings in a managed execution environment, a prompt update is strongly recommended.

New Features and Fixes in v2.1.126

The subsequent v2.1.126 (released May 1) improved the model selection screen to fetch a list of models from custom gateways, enabling correct listing of available models in configurations routed through internal proxies, Bedrock, or Vertex. Additionally, a new command claude project purge was added for bulk deletion of project state, allowing cleanup of workspaces duplicated for testing to be completed in a single command.

Stability Improvements

On the stability front, fixes include: (1) preventing session corruption from pasting large images via automatic downscaling, (2) fixing connection timeout errors after macOS sleep, and (3) fixing garbled display of Japanese, Korean, and Chinese filenames and output on Windows. In particular, the issue of sessions dying after long macOS sleep periods had accumulated multiple user reports as a known bug, and this fix stabilizes the typical scenario of resuming work after a break. To apply the update, confirm your version with claude --version, then apply via your channel such as npm i -g @anthropic-ai/claude-code@latest.

Impact: ★

Source: Claude Code Changelog

What to Watch Next — Key Developments Around Anthropic

The White House's opposition is the first major signal in the regulatory debate over AI use in cybersecurity. The next points to watch are how Anthropic modifies its Mythos deployment strategy and whether competing major AI companies face similar restrictions. Also worth tracking: whether the official announcement of a $50 billion raise comes after the May board meeting, whether 1M context window support expands to additional models, and whether further fixes are made to Claude Code's security settings.

3 Actions Developers Should Take This Week

Three actions are recommended in the short term. First, immediately switch any code using 1M tokens on Sonnet 4.5 / Sonnet 4 to Sonnet 4.6 / Opus 4.6. Second, teams running Claude Code in enterprise environments should update to v2.1.126 immediately and re-verify that allowManagedDomainsOnly-type settings are functioning as intended. Third, security teams evaluating Mythos-related features should continue monitoring for the latest information on distribution restrictions and update the premises of their contracts and SOC integration plans accordingly. The Clauder Navi editorial team will provide follow-up coverage in the next update.

Sources Cited in This Article

• TechCrunch: After dissing Anthropic for limiting Mythos, OpenAI restricts access to cyber too (2026-04-30)
• Anthropic Platform Release Notes
• TechCrunch: Sources: Anthropic could raise a new $50B round at a valuation of $900B (2026-04-29)
• Claude Code Changelog