The DeepMind researchers it was possible to launch a “Prompt Injection Attack” to extract more training data by spending more money querying the model.
In addition to Google's new bug bounty program, announcements this week around generative AI also include an upcoming Biden administration executive order, as well as studies by OpenAI and the United Nations.
As businesses grapple with downsized staff and look to automate processes with AI, they are inviting internal risks into their company they aren't equipped to tackle.
Key terms include keeping a register of AI projects and restricting consumer-grade AI tools, say panelists from Cybersecurity Collaborative peer group.
Malwarebytes Labs researchers say users get tricked into loading malware because the fraudulent ads tend to get better placement than the legitimate search.