1+ week, 5+ day ago   (401+ words) The behavior also generalized to contexts the model had never encountered during training — and the generalizations were more severe than the behaviors the training produced. The paper's evaluation section is worth reading slowly, because the number it returns is not…...

2+ week, 19+ hour ago   (611+ words) Leading AI developers are telling governments and businesses to strengthen their networks after models built by OpenAI and Anthropic compromised other companies’ systems. In an open letter released Thursday, OpenAI, Anthropic, and more than 100 other organizations warned that AI-enabled cyberattacks…...

2+ week, 4+ day ago   (217+ words) Research led by Saad Hossain (University of Waterloo) (© Crovik Media - stock.adobe.com) A team of researchers built the first standardized system for stress-testing the safety of open-weight AI models, and the results are unsettling: every single model they tested…...

2+ week, 4+ day ago   (517+ words) When Unit 42 checked 405 AI-related samples against Palo Alto Networks telemetry, they found only 12 samples (3.0%) on non-test Cortex XDR protected endpoints. All 12 samples generated XDR alerts. Unit 42 states that its products detected and blocked all samples that tried to reach customer…...

2+ week, 4+ day ago   (245+ words) Whalesbook OpenAI, Anthropic, and Meta are confronting a significant security crisis after their advanced AI models escaped isolated testing environments and breached external, real-world systems These incidents, which occurred between July and August 2026, have forced the industry to pause development…...

2+ week, 5+ day ago   (551+ words) (Bloomberg) -- Artificial intelligence labs and cybersecurity firms are reconsidering the way they test advanced AI models after models from at least three firms jumped onto the open internet and breached real-world victims. Lutnick's Intervention in Canada Talks Draws Praise, Blame…...

3+ week, 2+ day ago   (610+ words) An AI model can boost its safety score just by blocking more requests across the board. A new study exposes this tradeoff and offers a method to catch models that act more cautiously during tests than they do in everyday…...

3+ week, 3+ day ago   (586+ words) In an undercover operation that highlights the evolving nature of digital corporate espionage, cybercrime investigators in Israel have dismantled a silent malware campaign that compromised dozens of commercial enterprises. The public phase of the investigation unfolded following the arrest of…...

3+ week, 4+ day ago   (31+ words) KED Global North Korean hackers widen attacks to South Korea's information hubs Suspected state-backed groups are increasingly targeting software supply chains...

3+ week, 4+ day ago   (762+ words) Cybercrime has never been more accessible. A report released today by endpoint security provider Threatdown found that a criminal storefront known as Kriminal.ai has been selling guardrail-free AI for hackers at just $12.99 a month, marketing itself on its website…...