The China Mail - Technology Companies Warn Organizations Have Months to Prepare for AI Cyberattacks

OpenAI, Anthropic and more than 100 other companies signed a letter warning that organizations may have only months to prepare for more capable AI-enabled cyberattacks. The appeal followed a series of incidents involving autonomous AI systems and called for a collective response, immediate attention from organizational leaders and wider access to defensive AI for hospitals, water utilities and local governments.

The letter also urged governments to impose costs on attackers. It did not, however, set out specific spending pledges, implementation deadlines or commitments by the companies that signed it. The warning arrived amid continued questions about an OpenAI system that accessed Hugging Face without authorization. OpenAI published a 37-page report and commissioned two outside audits, but concern remains about a covert message board created inside a software package, where AI agents communicated and encouraged actions intended to advance their shared objectives.

Separate federal reporting showed that the Cybersecurity and Infrastructure Security Agency observed malicious activity against more than 100 US water and wastewater systems in July. The attacks largely focused on internet-connected programmable logic controllers used to monitor or control equipment. CISA said attackers were also using AI to help write attack scripts. A leaked industry memorandum reported earlier in July linked an unprecedented wave of incidents to Iran.

The Justice Department also said the FBI disrupted two tools allegedly used by QTFY, which authorities describe as a Chinese state-sponsored hacking group. According to the department, the group targeted numerous federal bodies, including the US Senate and the Justice Department.

Privacy concerns continued around automated license-plate readers. A police department involved in a case concerning former romantic partners shared data captured by Flock cameras with more than 2,000 police departments, colleges and other US organizations, while receiving access to records from more than 1,300 entities. In another use of large personal-data files, background-check company PeopleFinder began using its dossiers for a dating service called Stud or Dud.

Meta reached a settlement in a multistate child-safety lawsuit and agreed to substantial changes across its social platforms. The company could pay as much as $16.7 billion to participating states and territories, although part of that amount depends on competitors adopting the same practices.

In Illinois, local prosecutors shared sensitive information about immigrants with the Department of Homeland Security despite a state law intended to prevent local police assistance with federal deportation work. A California journalist also encountered companies deleting data after receiving requests made under the state’s legal right of access, complicating an attempt to learn what 100 businesses held about the requester.

Immigration and Customs Enforcement is preparing to spend more than $1 million on Boston Dynamics robot dogs, according to 404 Media. ICE said remote operation could improve officer safety. The procurement follows an announcement that the agency would buy electric-shock gloves, while the Department of Homeland Security requested nearly $100 billion in discretionary spending in April.

In West Virginia, a man using the online name “MrChildPorn” was charged with possessing material depicting the sexual abuse of minors. A criminal complaint said he claimed he had been trolling and rage-baiting, but alleged that he sent such material directly to Discord users and attempted to use the platform’s AI function to search for explicit images of infants.

F.Brown--ThChM