OpenAI's Cyber Restrictions: A Tale of Double Standards in AI Safety

In a striking turn of events, OpenAI has implemented access restrictions on its advanced AI system 'Cyber,' despite having previously criticized Anthropic for similar limitations on its 'Mythos' platform. This move has sparked debate about consistency in AI governance and the growing tension between safety measures and open access. Below, we explore key questions surrounding this development.

1. What exactly happened between OpenAI and Anthropic regarding access restrictions?

Earlier this year, OpenAI publicly criticized Anthropic for restricting user access to its Mythos AI model, arguing that such limitations undermined the principles of open research and user autonomy. However, in late April 2026, OpenAI quietly introduced similar restrictions on its own Cyber system, citing safety and misuse concerns. Critics note the apparent hypocrisy, as OpenAI's earlier stance was seen as a competitive jab. The restrictions include API rate limiting, content filtering, and requiring higher-tier subscriptions for certain capabilities.

2. Why did OpenAI originally criticize Anthropic's restrictions on Mythos?

OpenAI argued that Anthropic's restrictions on Mythos were overly cautious, potentially hampering innovation and user freedom. They claimed that AI safety should be achieved through transparency and user education rather than heavy-handed access controls. This public criticism positioned OpenAI as a champion of open access, making their subsequent restrictions on Cyber all the more controversial. Industry observers suggest OpenAI may have been motivated by competitive pressures, hoping to lure Mythos users to their platform by offering fewer limitations.

3. What is Cyber, and why did OpenAI choose to restrict access to it?

Cyber is OpenAI's latest AI model, designed for high-level reasoning and code generation. It gained rapid adoption in enterprise settings. According to an internal memo, OpenAI restricted access after detecting anomalous usage patterns, including attempted reverse engineering and attempts to generate harmful outputs. The company claims these measures are temporary and necessary to prevent malicious exploitation. However, the timing—coming shortly after their critique of Anthropic—has led many to question their motives.

4. How do the restrictions on Cyber compare to those OpenAI condemned on Mythos?

Both sets of restrictions involve limiting API throughput, blocking certain queries, and requiring higher authentication levels for sensitive tasks. However, OpenAI's current restrictions are reportedly more nuanced, phasing in only for high-volume users. In contrast, Anthropic's Mythos restrictions were broader and applied from launch. Despite these differences, the core principle—restricting user access based on safety risk—remains the same. This similarity is at the heart of the criticism that OpenAI is inconsistent.

5. How have the AI community and competitors responded to this development?

Reactions have been mixed. Some researchers accuse OpenAI of hypocrisy, pointing out that restrictions are inevitable for safety and that OpenAI's earlier stance was merely marketing. Others defend OpenAI, arguing that adaptive safety measures are necessary as threat landscapes evolve. Anthropic has declined to comment directly but issued a statement emphasizing that 'responsible AI governance requires principled, not reactionary, policies.' Meanwhile, smaller AI startups see this as an opportunity to promote their own access policies.

6. What are the broader implications for AI governance and public trust?

This incident highlights the difficulty of balancing openness with safety. If leading AI labs inconsistently apply restrictions, public trust may erode. It also raises questions about whether safety measures serve as genuine protection or competitive tools. Experts suggest that the industry needs transparent, collaborative guidelines for when and how to restrict access. Without such standards, similar controversies will likely recur, undermining the credibility of all AI safety efforts.

7. Could there be regulatory consequences for OpenAI's actions?

Potentially, yes. Regulators in both the EU and the US have been scrutinizing AI companies' access policies. OpenAI's reversal may prompt investigations into whether restrictions are used to stifle competition. If found to be anticompetitive, fines or forced changes could follow. However, given the safety rationale, regulators may focus on ensuring that restrictions are proportionate and justified. The outcome could set a precedent for how future AI access debates are framed.