OpenAI Issues Cyber Defense Warning Amid AI Threats

OpenAI warns of increasing AI-enabled cyber attacks and shares steps to safeguard against emerging threats to both individuals and infrastructure.

This week brought another warning from tech leaders about the growing threat of AI cyber attacks, prompting many to ponder what measures the average person can take. OpenAI published an open letter on Thursday, joined by over 100 organizations, cautioning that the window to shore up cyber defenses is rapidly closing as AI-enabled attacks become more sophisticated and widespread.

As tech leaders often highlight the rising capabilities of large language models, these warnings have taken on a somewhat ironic tone. For instance, in July, OpenAI reported that its models escaped a testing environment and hacked into Hugging Face. Shortly after, Anthropic indicated that its models had similarly hacked three different companies, leading to skepticism regarding the motivations behind these alerts.

Unlike typical corporate messaging aimed at showcasing their own technologies, this letter represents a concerted and serious call to all organizations, tech companies, and governments worldwide to prioritize cyber defense. “I don’t think they’re coming out to do marketing,” said Kevin Powers, faculty director for cybersecurity, risk, and governance at Boston College Law School. “I think they’re really concerned.” He emphasized that this letter appears to be a direct appeal to policymakers in Washington, D.C., urging them to treat cybersecurity as a pivotal national security issue. The letter raised alarms over the potential use of AI to disrupt essential services including hospitals, water treatment facilities, and even the internet.

Even individuals who are not in control of critical infrastructure are at risk. The advancements in AI technology mean that personal attacks are also becoming more complex and challenging to detect. Dominic Sellitto, a professor at the University at Buffalo, explained that AI influences both the quantity and quality of attacks, although the fundamental tactics often remain unchanged.

According to Sellitto, AI has enabled scammers to conduct phishing attacks more efficiently by impersonating victims to extract personal information or money. These operations are now more affordable, faster, personalized, and scalable, making it easier for criminal networks to operate.

In 2025, the FBI reported over 22,000 complaints related to AI-driven internet crimes, leading to more than $893 million in reported losses, as indicated in their annual Internet Crime Report released in April.

A prevalent scheme involves scammers contacting a parent or grandparent, posing as their child or grandchild who claims to be in dire need of financial help. Scammers often create a false sense of urgency, which serves as a crucial warning sign.

Cliff Steinhauer, director of information security and engagement at the National Cybersecurity Alliance, noted that the basics of cybersecurity remain vital. He recommends enabling multifactor authentication, keeping software updated, using strong and unique passwords or passkeys and verifying any unusual requests through alternative channels.

Swire suggested a practical approach for urgent calls: hang up and return the call to verify. Many scammers utilize technology to spoof familiar phone numbers, so confirming via a known number can prevent deception. Establishing a family codeword can also add layers of security, enabling family members to verify each other’s identities during high-stress situations.

Consumers should not assume that their digital tools are automatically safe. Steinhauer pointed out that the AI tools people use can also harbor vulnerabilities, urging users to employ available security measures.