OpenAI Reverses Course on Searchable ChatGPT Conversations
OpenAI has swiftly reversed a recent feature that allowed users to make their private conversations with ChatGPT discoverable by search engines, citing privacy concerns. The rollback comes amidst growing scrutiny over data privacy in AI interactions and recent revelations that shared ChatGPT conversations were being indexed by Google.
The decision was announced by OpenAI’s chief information security officer, who stated the company is rolling back the feature to address user privacy concerns. This move follows reports that Google was indexing shared ChatGPT links, making private exchanges, potentially containing sensitive personal or business information, publicly searchable. Users who utilized ChatGPT’s “Share” feature to generate unique URLs for conversations, often with the intention of sharing with a limited audience, found that these links could be discovered through general web searches, exposing prompts and responses.
This development underscores the ongoing challenges of data privacy in the rapidly evolving AI landscape. While OpenAI had previously introduced a “ChatGPT search” feature in November 2024, designed to allow ChatGPT to search the web for information and provide answers with sources, the issue at hand specifically relates to the searchability of private user conversations that were shared.
OpenAI CEO Sam Altman has also recently cautioned users that conversations with ChatGPT are not legally privileged, unlike discussions with medical or legal professionals, and could potentially be subject to disclosure in legal proceedings. This warning, coupled with the recent searchability issue, highlights a significant gap in user expectations regarding the confidentiality of their interactions with AI models. Many users, particularly younger individuals, have reportedly been using ChatGPT for highly personal matters, including therapy and life advice, under the assumption of privacy.
OpenAI has been actively working on strengthening its security initiatives. In October 2024, Dane Stuckey, formerly the Chief Information Security Officer at Palantir Technologies, joined OpenAI as its new CISO, working alongside Matt Knight, who was previously promoted to Head of Security. Stuckey emphasized that security is central to OpenAI’s mission of developing safe artificial general intelligence and maintaining high standards of compliance and trust. The company also states that its API, ChatGPT Enterprise, ChatGPT Team, and ChatGPT Edu products are covered by SOC 2 Type 2 reports, indicating adherence to industry standards for security and confidentiality.
Despite these security measures, the incident with searchable private conversations demonstrates the complexities of managing user data in AI systems that are increasingly integrated into daily life. Users are advised to exercise caution and consider any shared ChatGPT conversation as potentially public, even if initially intended for a small audience.