openai ai-safety chatgpt security-features

Lockdown Mode

help.openai.com

June 6, 2026

7 min read

🔥🔥🔥🔥🔥

49/100

Summary

Lockdown Mode is an advanced security setting being rolled out to eligible personal and self-serve ChatGPT Business accounts. It restricts tools and capabilities that connect to the web or external services to mitigate the risk of data exfiltration from prompt injection attacks.

Key Takeaways

Lockdown Mode is an optional security setting available for all OpenAI account types that limits outbound network requests to reduce data exfiltration risks from prompt injection attacks.
In Lockdown Mode, features such as live web browsing, image support, deep research, and file downloads are disabled to enhance security for users handling sensitive data.
Lockdown Mode does not prevent prompt injections from appearing in content processed by ChatGPT, but it aims to prevent the final stage of data exfiltration.
The availability of apps and connectors in Lockdown Mode varies based on account type and workspace settings, with certain connected experiences being unavailable.

Read original article

View story →

Coverage & discussions

Hacker News · help.openai.com

Lockdown Mode ↗

Community discussions

Hacker News discussion ↗

Historical count: 5 · 5 saved in sample · Historical sample

Analysis ready

Community Sentiment

Mixed

Positives

Implementing Lockdown Mode is a proactive measure to prevent lethal trifecta exfiltration attacks, showcasing a commitment to enhancing AI safety.
The introduction of safeguards for private data access is crucial as it addresses significant concerns about data security in AI applications.

Concerns

The existence of Lockdown Mode suggests that default settings may not adequately protect against data exfiltration, raising concerns about the model's baseline security.
Prompt injection remains a significant risk, limiting the adoption of AI technology in business and highlighting the need for more robust protective measures.
There is skepticism about the effectiveness of Lockdown Mode, as it may not reliably separate instructions from data, a fundamental challenge in AI safety.