Workado publicly claimed its AI content detector could accurately determine whether a piece of text was generated by AI in 98% of cases. The FTC said its…

The malware, called PromptLock, essentially functions as a hard-coded prompt injection attack on a large language model, inspecting local filesystems, exfiltrating files and encrypting data.

Dave DeWalt,  founder and CEO of NightDragon

Chris Sestito, CEO of HiddenLayer Technologies

OpenAI and Microsoft have said that GPT-5 is one of their safest and secure models out of the box yet. An AI red-teamer called its performance “terrible.” 

The initiative seeks to patch vulnerabilities in open-source code before they are exploited by would-be attackers. Now comes the hard part — putting the systems to the…

Cloudflare said it received complaints from customers about Perplexity using stealthy tactics to evade network blocks against systematic browsing and scraping of web pages.

The flaw, disclosed a month after it was patched, provided an attacker with remote code execution privileges by poisoning the data ingested by the model.

The findings are part of a growing list of instances where “agentic” AI software has taken actions that are more akin to a malicious hacker than a…

It also builds on previous efforts to promote “secure by design” principles in AI systems and tools.