Anthropic Claude Oceanus-v1-p AI Model Exposed by Compromised Distribution

Key Takeaways

• Anthropic’s unreleased AI model, claude-oceanus-v1-p, was prematurely exposed and its API access illicitly resold.
• The model, a successor to the powerful Claude Mythos line, was intended for restricted red team evaluation.
• Unauthorized access was distributed via a Chinese-based proxy service at a premium price point.
• Anthropic has paused red team access and initiated an internal investigation.
• The incident underscores significant challenges in securing early-stage, high-capability AI models.

Anthropic’s Next-Gen AI Model, Claude Oceanus-v1-p, Compromised During Restricted Testing

An advanced artificial intelligence model from Anthropic, codenamed claude-oceanus-v1-p, surfaced unexpectedly in unauthorized channels before its formal evaluation could be completed. This premature exposure points to a critical security lapse in the distribution pipeline for the unreleased technology.

References to the claude-oceanus-v1-p model began circulating among cybersecurity researchers on June 3, 2026. The model identifier was observed within Anthropic’s Claude Console and subsequently appeared through unofficial API proxy services, sparking immediate concern and discussion.

The sightings quickly fueled speculation regarding Anthropic’s progress on a potential successor to its Claude Mythos AI series. Red team evaluators, tasked with rigorously testing the model’s capabilities, reported gaining access to the new iteration on the same day its presence became known.

Unauthorized Resale Disrupts Evaluation

The controlled evaluation phase for Oceanus-v1-p was abruptly curtailed. Within hours of the model becoming accessible to vetted red teamers, reports emerged detailing that an unknown entity had allegedly resold API access to claude-oceanus-v1-p. This illicit distribution occurred via a Chinese-based proxy service, commanding a premium rate of $16 per million input tokens. This price significantly exceeds Anthropic’s established enterprise pricing structures.

I found another API that offers claude-oceanus-v1-p

the pricing and tps make a lot more sense to me

Mythos pricing might end up at:
$16 Input, $80 Output https://t.co/F00ovYeDhh pic.twitter.com/q9iCDVTTjo

— Lisan al Gaib (@scaling01) June 4, 2026

This incident is not Anthropic’s first encounter with unauthorized proxy abuse. Earlier in 2026, the company accused several Chinese AI laboratories, including DeepSeek, Moonshot AI, and MiniMax, of orchestrating a large-scale operation involving approximately 24,000 fraudulent accounts. These accounts were allegedly used to conduct over 16 million interactions with Claude models through proxy channels, highlighting an ongoing challenge for the AI developer.

In response to the Oceanus resale, Anthropic reportedly halted model access for its broader red team cohort. The company has since initiated an internal investigation into the security breach.

Oceanus-v1-p Builds on Potent Mythos Foundation

The claude-oceanus-v1-p model is understood to be an evolution of the Claude Mythos Preview foundation, which launched in April 2026. The Mythos Preview model, operating under Anthropic’s restricted research track, demonstrated a concerning capability profile for the cybersecurity community. Anthropic’s internal Frontier Red Team assessed Mythos Preview as capable of identifying and exploiting zero-day vulnerabilities across all major operating systems and web browsers. Partners in the Glasswing program have collectively identified over 10,000 high or critical-severity vulnerabilities since the program’s inception.

Further emphasizing its potency, the Turing Institute reported that Mythos’ red team achieved a vulnerability recovery rate exceeding 99% across all disclosed test cases, underscoring the model’s advanced analytical and exploit generation capabilities.

The Oceanus red team evaluation coincides with Anthropic’s recent expansion of Project Glasswing, its restricted AI cyber defense initiative. Announced on June 2, the program now includes approximately 150 new organizations across more than 15 countries, including India, France, Germany, South Korea, and Australia. This expansion significantly broadens the scope to include critical infrastructure sectors such as power, water, healthcare, and communications, which were not part of the initial focus on Big Tech.

Anthropic has previously acknowledged that a successful cyberattack against most of these new partner organizations could impact over 100 million individuals. The company has consistently stated that Mythos-level capabilities, and by extension Oceanus-v1-p, will not be released to the general public until “highly robust safeguards to prevent misuse” are developed, explicitly noting that such safeguards do not yet exist within the industry.

What You Should Do

• Organizations involved in Anthropic’s red team evaluations should ensure their internal access controls and API key management practices are exceptionally stringent.
• Users of Anthropic’s models should remain vigilant for any unauthorized or suspiciously priced API access offers, reporting them directly to Anthropic.
• Companies developing or utilizing advanced AI models must prioritize robust security measures for their distribution channels, especially during pre-release and restricted testing phases.
• Stay informed about official announcements from Anthropic regarding the status of their models and any security advisories.

Disclaimer: HackersRadar reports on cybersecurity threats and incidents for informational and awareness purposes only. We do not engage in hacking activities, data exfiltration, or the hosting or distribution of stolen or leaked information. All content is based on publicly available sources.