OpenAI has launched Lockdown Mode, a new security feature aimed at protecting users who handle sensitive information from prompt injection attacks. The company rolled out the feature on June
OpenAI has launched Lockdown Mode, a new security feature aimed at protecting users who handle sensitive information from prompt injection attacks. The company rolled out the feature on June 7, 2026.
According to The Economic Times, Lockdown Mode is designed to defend against prompt injection, an attack method in which malicious instructions embedded in inputs attempt to override or manipulate the AI model's behavior.
Prompt injection is one of the most frequently discussed vulnerabilities in large language model deployment. Attackers craft inputs that instruct the model to ignore prior instructions or expose restricted information. Lockdown Mode applies additional controls to reduce that risk.
OpenAI has not published a full technical breakdown of Lockdown Mode's implementation in the current rollout. The company described the feature as targeted at users managing sensitive data, suggesting an audience that includes enterprise clients, researchers, and security-conscious professionals.
The launch reflects a pattern at OpenAI of adding access tiers and security layers as the platform handles increasingly high-stakes use cases. Enterprise deployments in legal, healthcare, and government sectors have heightened the need for input-level security controls.
Also Read: Two AI Rivals, One Compute Bill: Inside Google's $30B SpaceX Move
OpenAI's push into safety-focused tooling follows a period of heightened scrutiny over AI model behavior. In the weeks prior to this launch, the heads of OpenAI, Anthropic, Google DeepMind, and Microsoft jointly urged Congress to mandate synthetic DNA screening, citing AI's potential role in biosecurity threats. That letter, reported in late May and early June 2026, placed AI safety at the center of Washington's legislative agenda.
The Lockdown Mode release arrives as competitors including Anthropic have also tightened access controls for their most capable models. Anthropic limited access to its Claude Mythos system earlier this year as part of a voluntary pre-release review framework.
OpenAI's security feature additions come ahead of the company's confidential IPO filing, submitted to the SEC on June 1. Demonstrating enterprise-grade security controls is likely to factor into investor confidence ahead of any public offering.
Read Next: Tencent Hires Former OpenAI Researcher Yao Shunyu As Chief AI Scientist In AGI Push
