SAN FRANCISCO - In a major shift in the global artificial intelligence landscape, OpenAI has announced a massive expansion of its Trusted Access for Cyber (TAC) program. The move aims to put the company’s most sophisticated reasoning models specifically the latest GPT-5.4-Cyber into the hands of all vetted levels of government, ranging from high-level federal agencies like the Pentagon to state and local offices.

The strategy represents a "democratization of defense," designed to tip the scales in favor of cybersecurity professionals as hackers increasingly leverage AI to automate sophisticated attacks.

OpenAI vs. Anthropic: Two Philosophies of Defense

The expansion highlights a growing divide between AI giants on how to handle "frontier" models that possess dual-use capabilities.

  • The OpenAI Approach: Under the leadership of Sasha Baker, head of national security policy, OpenAI argues that keeping powerful tools restricted to elite corporations (the "Fortune 50") leaves the rest of society vulnerable. Their "Action Plan for the Intelligence Age" suggests that broader access for vetted defenders is the only way to outpace malicious actors.

  • The Anthropic Contrast: Rival firm Anthropic has taken a more restrictive stance with its latest model, Mythos. Capable of identifying 27-year-old "zero-day" vulnerabilities in seconds, Mythos has been deemed too dangerous for public release. Instead, it is being deployed via Project Glasswing, a tightly controlled consortium involving a small group of partners like AWS, Microsoft, and the Department of Defense.

Anthropic believes a "slower, more cautious" rollout prevents an AI-fueled arms race, whereas OpenAI contends that the arms race has already begun and defenders need every advantage immediately.

The White House Intervention

This corporate tension has reached the highest levels of the U.S. government. Just this week, representatives from OpenAI, Anthropic, Google, and major financial institutions met at the White House with the National Cyber Director. The meeting, which followed an OpenAI-led workshop with the Department of Homeland Security and DARPA, focused on how AI companies can "show up" to protect national infrastructure.

Sources indicate that the Trump administration is now reconsidering its deregulatory stance on AI. While the administration previously rescinded safety evaluation requirements, the sheer power of models like Mythos and GPT-5.4-Cyber has led officials to discuss a new executive order. This could establish a "working group" of tech executives and government officials to oversee model releases before they reach the public, aiming to prevent a devastating AI-enabled cyberattack.

Proactive Security for the "Intelligence Age"

As part of its new roadmap, OpenAI is committing $10 million in API credits to accelerate cyber defense through its Cybersecurity Grant Program. Beyond government applications, the company plans to introduce:

    •    Advanced Account Security: New features for everyday ChatGPT users to improve personal cyber hygiene.

    •    Codex Security Agents: Autonomous AI tools designed to find, validate, and propose fixes for software vulnerabilities in real-time.

    •    Intelligence Sharing: A framework for public-private coordination to identify global threats with greater precision.

While Sasha Baker admits the latest AI capabilities are a "wake-up call" for the community, she maintains that there is no need for panic—provided that the transition to AI-driven defense is handled with efficiency and urgency.