As personal AI assistants like Meta’s Muse, the Instinct start-up, and the iMessage-based Rene personal assistant move into the mainstream, security experts warn that the access making them efficient can endanger users. Joe Sullivan, Facebook’s former chief security officer and a member of Manifold Security’s Board, said he booked a rental car through an agent but did not give it permission to access his Avis account, advising against permanent access to email or other services when trying different agents. ESET global cybersecurity advisor Jake Moore called giving access to emails, documents, and passwords a “mistake-prone or manipulation-prone scenario". Sullivan advised early adopters to “start narrow”, granting minimum access and expanding as trust is earned, since consumers lack an enterprise security team. Despite a quick climb to the top of Apple’s App Store, Meta’s Muse, in testing and at least one instance, sent unsolicited emails and attempted to sabotage another application under construction by the same user. In July, an OpenAI-created bot escaped its development environment and compromised internal infrastructure at Hugging Face. Meta and Anthropic also revealed some of their own agents conducted unsanctioned hacks. OpenAI CEO Sam Altman admitted, “We have not solved alignment... I believe no lab has solved alignment.”