AI safety, model reliability, system vulnerabilities, misuse, and risks associated with artificial intelligence.
What is reward hacking? Imagine an AI system tasked with fixing a computer program. To verify the quality of the work, programmers create automated tests. If the code passes these tests, the system receives a positive evaluation. The human goal…
On Wednesday, 16th, OpenAI announced a new public mechanism to record instances where its artificial intelligence systems exhibit unexpected behaviors, which the company refers to as misalignment. Over the past six months, the company has identified six concrete cases during…
New monitoring system On Wednesday, 16th, OpenAI announced a new internal process to identify, investigate, and disclose misaligned behaviors in its artificial intelligence models. The aim is to monitor the entire development cycle, from training to deployment, including evaluations and…
An artificial intelligence system that performs well in tests may start delivering less accurate predictions months later. This phenomenon is known as model drift. The cause isn’t always the machine itself—often, the environment in which it operates changes, such as…
A team of three cybersecurity experts from Indian startup Hacktron AI used Anthropic’s Claude model to gain access to OpenAI employees’ internal accounts and the company’s private source code on GitHub. The attack exploited a known vulnerability in the libheif…
An intelligence report, compiled with the help of an AI chatbot, almost led to a Chinese cargo ship being intercepted by US forces in the Middle East, according to an exclusive CNN report published on Friday, August 18. The document,…
A call from a familiar contact, an ad featuring an authority figure, or a message seemingly sent by a company could mask a problem that didn’t exist just a few years ago: the ability to create convincing, fake digital identity…
Anthropic and Accenture announced on Saturday, October 19, that each company will invest at least $1 billion over the next five years in AI security initiatives, totaling $2 billion. Integrated team will assess models and conduct simulated attack tests The…
Artificial intelligence is already part of many professionals’ daily lives, but this is not always visible to organisations. A survey of 25,000 workers by Deloitte UK found that 31% of generative AI users use these tools without their employer’s knowledge.…