Anthropic Secures Government Reinstatement Amid Operational Challenges

Anthropic's AI models are reinstated for U.S. government use after a court order, but the firm faces security incidents and platform capacity challenges.

Anthropic has achieved a significant legal victory with U.S. federal authorities reversing a ban on its artificial intelligence models. This regulatory win, however, arrives as the AI startup contends with parallel pressures from security incidents and platform capacity constraints.

Security Incidents Draw Scrutiny

Even as it celebrates its legal success, Anthropic's security practices are facing increased examination. The AI firm Mercor confirmed yesterday that a vulnerability at the service provider LiteLLM led to unauthorized data access affecting Anthropic's systems. Intruders were able to view system logs and login credentials.

This follows a separate March incident where proprietary source code was accidentally exposed through the public release of an npm package. In response, Anthropic has filed over 8,000 takedown requests to prevent the dissemination of details concerning internal projects, including autonomous assistants. These consecutive breaches have prompted enterprise clients to reevaluate the risks of over-reliance on individual AI providers.

Court Order Forces Regulatory Reversal

The reinstatement stems from a preliminary injunction issued by a California district court on March 26. In compliance, the U.S. General Services Administration (GSA) confirmed on Friday that Anthropic's technologies are once again permitted for use within its internal systems. The agency was compelled to rescind a restrictive policy enacted in February and revert to the prior status quo.

This reintegration covers central procurement channels such as GSA Chat and the Multiple Award Schedule (MAS), which federal agencies use to obtain software. For Anthropic, the decision provides substantial regulatory relief, legally safeguarding the integration of its AI tools into government workflows.

Platform Strain and Advanced Model Testing

Operational hurdles are also emerging on Anthropic's user platform. Customers of "Claude Code" are exhausting their daily session allowances much faster than anticipated due to rising token usage during peak times. To manage the load, the company has implemented capacity restrictions between 5:00 a.m. and 11:00 a.m. Pacific Time.

Concurrently, an internal model dubbed "Mythos" is attracting attention from cybersecurity analysts. Reports indicate Anthropic is evaluating the model's advanced logical capabilities for dual-use potential. The technology could be applied defensively or, alternatively, to accelerate the discovery of software vulnerabilities.

While the GSA's renewed approval grants Anthropic fresh opportunities within the public sector, it simultaneously forces the company to undertake significant improvements to the stability and security of its core infrastructure.