TypeSafe’s Jev ‘Even More Susceptible to Attack’ than LLMs, Says Security Vendor

Decision Models Like Jev Fall to Prompt Tricks Despite No-Hallucination Claims

HiddenLayer researchers tricked TypeSafe AI's Jev and similar decision models with indirect prompt injection in a simulated checkout attack. Despite claims of no hallucinations or type errors, malicious web ads shifted click probabilities and hijacked agent actions. Recent tests from Check Point, Opper.ai and Lasso confirm the vulnerability persists across providers. Structured outputs constrain answers but do not eliminate steering risks.

TypeSafe’s Jev ‘even more susceptible to attack’ than LLMs, says security vendor

Although the Jev model rolled out last month by TypeSafe claims to have significant advantages over traditional LLMs, a security report published Wednesday argued that its streamlined approach makes it more vulnerable to some kinds of attack, compared to traditional LLMs.   “In some cases, [Jev] is even more vulnerable to attacks, with both adversarial text attacks and prompt injection being able to disrupt the model’s outputs, resulting in inco…

Jev and other decision models can still be tricked with prompt injection

Never making type errors does not mean never making errors

TypeSafe’s Jev ‘even more susceptible to attack’ than LLMs, says security vendor #AI - National Cyber Security Consulting

LLM, or something else? There seems to be debate about whether Jev is technically a large language model (LLM), with TypeSafe suggesting it is not, but HiddenLayer saying that it appears to be one. TypeSafe has even coined a name for the category: System One, which TypeSafe described as “a new class of frontier models […] Thank you for subscribing to our RSS feed!