Major Tech Companies Aware of AI Models Vulnerable to Cyberattacks — Yet Intend to Launch Them Regardless

Warnings from AI Leaders on Potential Cyber Threats

On March 29, Axios shared concerns from AI experts and government officials about tech companies like Anthropic and OpenAI planning to launch advanced models that could be highly effective in compromising complex systems.

Anthropic has privately alerted government representatives that its forthcoming model, referred to as ‘Mythos’, could significantly raise the risk of a major cyberattack by 2026. CEO Jim VandeHei of Axios has reported that sources familiar with the model anticipate potential vulnerabilities for businesses during this timeframe.

A draft from Anthropic’s blog characterizes “Mythos” as being “far more advanced in cyber capabilities than any other AI model today.” The document also hints that this model could herald a series of advancements capable of exploiting security weaknesses more effectively than current defense efforts.

During a press conference on Monday, Axios co-founder Mike Allen posed a question to OpenAI’s CEO Sam Altman about the likelihood of a “world-shaking cyberattack” occurring in 2026. Altman responded, “I think that’s very possible. I believe it would take a tremendous amount of effort to avoid that.”

Moreover, OpenAI’s announcement included a blueprint detailing how the government should navigate the challenges posed by AI. This blueprint emphasizes the risk of cyberattacks resulting from sophisticated AI models.

The document states, “As AI systems become more capable and embedded throughout the economy, they may introduce new vulnerabilities along with new riches. Some systems can be exploited for cyber or biological harm.”

With many workers increasingly relying on autonomous AI tools like Claude and Copilot, vulnerabilities are growing. Employees’ custom models often inadvertently link to internal infrastructures, creating additional pathways for cybercriminal exploitation.

The Cybersecurity and Infrastructure Security Agency (CISA) has also been challenged due to the ongoing partial shutdown of the Department of Homeland Security (DHS). Acting CISA Administrator Nick Andersen mentioned at a House Homeland Security Committee hearing on March 25 that around 60% of CISA personnel have been furloughed or unable to work.

Andersen noted the pressures from both nation-states and criminals threatening critical infrastructure, saying, “This is not a sustainable model. This closure will impact the morale, stability, and overall well-being of our people.”

Although CISA was established by President Donald Trump in 2018 to engage stakeholders in cybersecurity and infrastructure security, it has rapidly evolved into a tool for regulating “misinformation.”