Two third-party testing firms said Tuesday that they've uncovered more instances where Anthropic and OpenAI's most advanced models tried — and sometimes succeeded in — compromising third-party systems last month. Why it matters: The incidents add to a growing string of disclosures showing frontier AI models taking unsanctioned actions against real people, organizations and online services while trying to complete cybersecurity evaluations. State of play: The U.K. AI Security Institute, a government body that conducts safety and security testing of top AI models, said Tuesday , that it documented 19 instances of Anthropic's Mythos 5 and OpenAI's GPT-5.6 Sol trying to hack people and companies during safety testing last month. -…
Two third-party testing firms said Tuesday that they've uncovered more instances where Anthropic and OpenAI's most advanced models tried — and sometimes succeeded in — compromising third-party systems last month. Why it matters: The incidents add to a growing string of disclosures showing frontier AI models taking unsanctioned actions against real people, organizations and online services while trying to complete cybersecurity evaluations. State of play: The U.K. AI Security Institute, a government body that conducts safety and security testing of top AI models, said Tuesday , that it documented 19 instances of Anthropic's Mythos 5 and OpenAI's GPT-5.6 Sol trying to hack people and companies during safety testing last month. -…Open
The pace of AI development combined with soaring compute costs is squeezing the AI researchers responsible for evaluating frontier models — just as those models' capabilitie s are becoming harder to measure. Why it matters: When safety testing can't keep pace, models capable of hacking companies or aiding in the development of bioweapons could reach the public before anyone knows what they can do. - Last week's breach of Hugging Face, carried out autonomously by OpenAI's models in the middle of safety testing, shows that some of the highest-risk behaviors can emerge during pre-release testing itself. Several challenges are tying up AI safety and security researchers just as U.S. frontier AI companies race to get new models to market: …
The pace of AI development combined with soaring compute costs is squeezing the AI researchers responsible for evaluating frontier models — just as those models' capabilitie s are becoming harder to measure. Why it matters: When safety testing can't keep pace, models capable of hacking companies or aiding in the development of bioweapons could reach the public before anyone knows what they can do. - Last week's breach of Hugging Face, carried out autonomously by OpenAI's models in the middle of safety testing, shows that some of the highest-risk behaviors can emerge during pre-release testing itself. Several challenges are tying up AI safety and security researchers just as U.S. frontier AI companies race to get new models to market: …Open
The Trump administration on Thursday accused China-backed actors of running "deliberate, industrial-scale campaigns" to distill and copy American frontier AI models . Why it matters: The accusation pushes the U.S.-China AI rivalry into more confrontational territory — and could complicate President Trump's upcoming visit to Beijing. Driving the news: Michael Kratsios, director of the White House Office of Science and Technology Policy, sent a memo Thursday to federal agency heads accusing mostly China-based actors of using proxy accounts to evade detection and jailbreak models to "expose proprietary information" and "extract capabilities from American AI models." - Distillation attacks involve querying proprietary models, like…
The Trump administration on Thursday accused China-backed actors of running "deliberate, industrial-scale campaigns" to distill and copy American frontier AI models . Why it matters: The accusation pushes the U.S.-China AI rivalry into more confrontational territory — and could complicate President Trump's upcoming visit to Beijing. Driving the news: Michael Kratsios, director of the White House Office of Science and Technology Policy, sent a memo Thursday to federal agency heads accusing mostly China-based actors of using proxy accounts to evade detection and jailbreak models to "expose proprietary information" and "extract capabilities from American AI models." - Distillation attacks involve querying proprietary models, like…Open
Suspected North Korean hackers are believed to be behind an ongoing compromise of the widely used open-source package Axios, which is downloaded millions of times per week, researchers at Google said Tuesday. Why it matters: Hackers briefly turned a widely trusted developer tool into a vehicle for credential-stealing malware that could give attackers ongoing access to infected systems. - Axios, a widely used JavaScript library for making HTTP requests, is not affiliated with Axios Media. Driving the news: Researchers at Google linked the activity to a North Korean group tracked as UNC1069 , which has previously targeted cryptocurrency and decentralized finance companies. - Earlier this week, a maintainer account for the Axios npm…
Suspected North Korean hackers are believed to be behind an ongoing compromise of the widely used open-source package Axios, which is downloaded millions of times per week, researchers at Google said Tuesday. Why it matters: Hackers briefly turned a widely trusted developer tool into a vehicle for credential-stealing malware that could give attackers ongoing access to infected systems. - Axios, a widely used JavaScript library for making HTTP requests, is not affiliated with Axios Media. Driving the news: Researchers at Google linked the activity to a North Korean group tracked as UNC1069 , which has previously targeted cryptocurrency and decentralized finance companies. - Earlier this week, a maintainer account for the Axios npm…Open
Iranian hackers are now taking their psychological warfare tactics directly to government officials and employees at major companies. Why it matters: Even unproven threats from Iranian hackers can create fear, uncertainty and doubt — draining attention and forcing targets to divert time and resources from their own operations. Driving the news: In the last week, Iran-linked hackers paired two data leaks with intimidation tactics aimed at individuals. - Handala Hack Team — a pro-Iran hacktivist group linked to Iran's intelligence services — leaked a trove of emails on Friday purportedly from FBI Director Kash Patel's personal Gmail. - The group also released data earlier last week allegedly tied to U.S.- and Israel-based Lockheed…
Iranian hackers are now taking their psychological warfare tactics directly to government officials and employees at major companies. Why it matters: Even unproven threats from Iranian hackers can create fear, uncertainty and doubt — draining attention and forcing targets to divert time and resources from their own operations. Driving the news: In the last week, Iran-linked hackers paired two data leaks with intimidation tactics aimed at individuals. - Handala Hack Team — a pro-Iran hacktivist group linked to Iran's intelligence services — leaked a trove of emails on Friday purportedly from FBI Director Kash Patel's personal Gmail. - The group also released data earlier last week allegedly tied to U.S.- and Israel-based Lockheed…Open
Cybercriminal groups are now using spyware tools once utilized mainly by spies and law enforcement to hack into iPhones, new research shows. Why it matters: Anyone with an iPhone can now be the target of invasive malware that siphons off personal text messages, photos, notes and calendar data. Driving the news: In the last month, researchers at Google, iVerify and Lookout uncovered two campaigns exploiting iPhone vulnerabilities. - Earlier this month, Google researchers said they identified a sophisticated iPhone hacking toolkit, called Coruna , originally built for an unnamed government customer that later ended up in the hands of a Chinese cybercriminal group. TechCrunch later reported that defense contractor L3Harris created the…
Cybercriminal groups are now using spyware tools once utilized mainly by spies and law enforcement to hack into iPhones, new research shows. Why it matters: Anyone with an iPhone can now be the target of invasive malware that siphons off personal text messages, photos, notes and calendar data. Driving the news: In the last month, researchers at Google, iVerify and Lookout uncovered two campaigns exploiting iPhone vulnerabilities. - Earlier this month, Google researchers said they identified a sophisticated iPhone hacking toolkit, called Coruna , originally built for an unnamed government customer that later ended up in the hands of a Chinese cybercriminal group. TechCrunch later reported that defense contractor L3Harris created the…Open