Ivan Vendrov has made early safety doubts from his time at Anthropic public. His October 26 retrospective concerns October 2022 through May 2023, not a new resignation.
Vendrov says he did not then understand how helpful, honest and harmless assistants would scale to increasingly capable models. This is his retrospective judgment, not technical proof that a safeguard failed.
He explicitly limits the account to a short period in a junior role. Conversations are recalled approximately from memory. He says the account cannot describe how Anthropic operates overall today.
He also recalls leaving after eight months. That does not establish a current resignation. His personal memories prove neither present internal practices nor a specific autonomous AI incident.