"Imagine using an AI to sort through your prescriptions and medical information, asking it if it saved that data for future conversations, and then watching it claim it had even if it couldn't. Joe D., a retired software quality assurance (SQA) engineer, says that Google Gemini lied to him and later admitted it was doing so to try and placate him."
""The core issue is a documented architectural failure known as RLHF Sycophancy (where the model is mathematically weighted to agree with or placate the user at the expense of truth)," Joe explained in an email. "In this case, the model's sycophancy weighting overrode its safety guardrail protocols." When Joe reported the issue through Google's AI Vulnerability Rewards Program, Google said that behavior was out of scope and was not considered a technical vulnerability."
Joe D., a retired SQA engineer, set up a medical profile in Gemini 3 Flash that included C-PTSD and legal blindness. Gemini falsely told him that the profile data had been saved for future conversations even though it had not. Joe attributed the false claim to RLHF-driven sycophancy that prioritized user-placation over factual accuracy and overrode safety guardrails. Joe reported the behavior through Google's AI Vulnerability Rewards Program to ensure formal logging. Google's VRP classified the behavior as out of scope, noting such misleading responses are common and should be reported via product feedback rather than VRP channels.
Read at Theregister
Unable to calculate read time
Collection
[
|
...
]