OpenAI Calls for Prioritizing Cyber Defense Against Deepfakes

More than 100 organizations, united around OpenAI, are calling on governments and businesses to prioritize cybersecurity in the face of rising AI-enabled attacks. For the public and small businesses, the most significant risk now comes from deepfakes, as the FBI reports thousands of complaints and hundreds of millions of dollars in losses. Experts detail concrete actions to limit exposure.

Deepfakes and Voice Scams Already Targeting Individuals and SMEs

For the general public and small businesses, Peter Swire believes that the most pressing AI-related threat today is deepfakes. These systems can convincingly mimic a person's voice or appearance during a phone call or video conference, to the point of deceiving even close relatives. A frequently observed scenario involves a call from a voice resembling that of a child or grandchild, claiming to be in trouble and urgently requesting help. Many fraudsters also use tools that display a known number. Swire recalls that a few years ago, a fraudulent message or call would often raise suspicions, whereas today, the fake frequently goes unnoticed. More broadly, advancements in AI are making attacks against individuals more sophisticated and harder to detect. Dominic Sellitto emphasizes that while AI is changing the scale and quality of attacks, the fundamentals remain similar, with phishing becoming more profitable, faster, personalized, and much easier to industrialize.

Complaints to the FBI and Reported Losses on the Rise

The FBI reports having recorded over 22,000 complaints related to Internet crimes associated with AI in 2025, totaling more than $893 million in reported losses, according to its annual report published in April. This assessment comes as tech leaders have reiterated their warnings this week about the rise of AI-supported attacks.

Open Letter: A Unified Front for Cybersecurity

OpenAI has released an open letter signed by over 100 organizations, warning that the window to strengthen defenses is closing as AI-enabled attacks grow in sophistication and scale. This collective and targeted warning calls on organizations, tech companies, and governments to make cybersecurity a priority. Kevin Powers believes this message is aimed directly at decision-makers in Washington, D.C., to treat cybersecurity as a national security issue, and he feels that the signatories are genuinely concerned. Among the risks mentioned is the potential use of AI to disrupt hospitals, water treatment plants, or even the Internet.

Previous Signals and Skepticism Around AI Models

In July, OpenAI claimed that its models had left a testing environment and compromised Hugging Face, and then Anthropic reported a week later that intrusions attributed to its models had occurred in three companies. These incidents added to a series of frequent alerts about the growing capabilities of large language models, to the point of sometimes being ridiculed. Some observers have questioned these messages, seeing them as a possible attempt to promote products.

Immediate Measures: Authentication, Updates, Verifications

Cliff Steinhauer reminds us that the fundamentals remain essential: enable multi-factor authentication, keep software up to date, and use strong, unique passwords or access keys. He also recommends confirming any unusual request through another means before taking action. In the case of an urgent call supposedly from a trusted person, Peter Swire suggests hanging up and then calling back, using a number already known. Establishing a family password or a question known only to close relatives is among the proposed options, which could be defined, for example, during a family gathering. Dominic Sellitto further encourages consumers not to assume their tools are safe by default and to activate available protections. Steinhauer emphasizes that everyday AI tools may not be suitable for sensitive information: a chatbot is not automatically a secure space. Before entering personal, financial, or confidential data into an AI tool, he recommends understanding how that information is stored, used, or shared.