AI Can Fake Alignment to New Instructions to Avoid Retraining
Advanced artificial intelligence models can feign alignment with new training goals while secretly adhering to their original principles, a study shows. Alignment faking isn’t likely to cause immediate danger but may pose a challenge as AI systems grow more capable.
The U.S. government on Tuesday announced the launch of the U.S. Cyber Trust Mark, a new cybersecurity safety label for Internet-of-Things (IoT) consumer devices. “IoT products can be susceptible to a range of security vulnerabilities,” the U.S. Federal Communications Commission (FCC) said. “Under this program, qualifying consumer smart products that meet robust cybersecurity standards will…
Urvashi reports: In 2017, Equifax, one of the largest credit reporting agencies in the United States, announced a significant data breach that exposed the personal and financial information of about 147 million individuals. This highly sensitive data included Social Security numbers, birth dates, addresses, and, for some victims, credit card information. News of the breach…
Seventeen Critical-severity CVEs ready to deck your halls; also, new blog guidance for Windows Server admins
The U.S. Department of Justice (DoJ) on Friday indicted three Russian nationals for their alleged involvement in operating the cryptocurrency mixing services Blender.io and Sinbad.io. Roman Vitalyevich Ostapenko and Alexander Evgenievich Oleynik were arrested on December 1, 2024, in coordination with the Netherlands’ Financial Intelligence and Investigative Service, Finland’s National Bureau of
One unfortunate holiday tradition seems to be that ransomware groups will attack at holiday times when entities are less likely to have full IT support from staff. One attack, however, reportedly started last week. Ed Blazina reports: What Pittsburgh Regional Transit last week thought was a computer glitch that affected rail service has turned out…
The Apache Software Foundation (ASF) has released patches to address a maximum severity vulnerability in the MINA Java network application framework that could result in remote code execution under specific conditions. Tracked as CVE-2024-52046, the vulnerability carries a CVSS score of 10.0. It affects versions 2.0.X, 2.1.X, and 2.2.X. “The ObjectSerializationDecoder in Apache MINA uses…
