#openai Tech News.

OpenAI has rolled out Lockdown Mode to eligible personal accounts and self-serve ChatGPT Business accounts. Lockdown Mode limits outbound network requests to prevent data exfiltration from prompt injection attacks. It does not prevent prompt injections from appearing in processed content. Lockdown Mode targets the exfiltration vectors to steal data, a key leg of the Lethal Trifecta. This mode uses deterministic mechanisms not evaluated by AI systems.

OpenAI has introduced Lockdown Mode, an optional security setting that limits features to protect users from prompt injection attacks. The feature is designed to offer advanced protection. Lockdown Mode is an optional setting, allowing users to choose their level of security. This setting is intended to provide users with more control over their security.

OpenAI CEO Sam Altman said his previous comments about AI's impact on jobs were 'pretty wrong'. He made the statement at the Commonwealth Bank of Australia conference in Sydney. Altman initially predicted significant social and economic implications from AI adoption. The reversal comes as AI adoption continues to rise.

Microsoft AI chief Mustafa Suleyman says the company was 'set free' from its partnership with OpenAI to pursue superintelligence, ending a three-year relationship that catapulted Copilot products into the enterprise mainstream and added hundreds of billions of dollars to its market capitalization.

OpenAI will voluntarily comply with President Trump's new AI executive order, which requests access to the latest AI models 30 days before their release. The order was originally drafted to require 90-day submissions. OpenAI will submit its next-gen AI models to a benchmarking process to assess advanced cyber capabilities. The company has also suggested ways for governments to track AI safety and security issues globally.

OpenAI has added email sending directly from ChatGPT on the web. Users can draft, tweak, and send emails within writing blocks without leaving the chat interface. Writing blocks were introduced late last year and now support inline edits and suggestions. ChatGPT for Excel and Google Sheets is no longer in beta and is available globally.

OpenAI released new AI tools for developers including code generation and debugging features. The tools support real-time code suggestions and error detection in multiple programming languages. Updates are available through the OpenAI Developer API. The tools aim to improve developer productivity with minimal setup.

Microsoft unveiled seven new in-house AI models at its Build conference. Its flagship MAI-Thinking-1 model was built from scratch without distillation. Using McKinsey benchmarks, it outperformed OpenAI's GPT 5-5 with 10 times better cost efficiency. The company introduced models for coding, image, voice, and transcription. Microsoft can reduce costs by cutting payments to third-party AI firms like OpenAI.

OpenAI has begun rolling out Lockdown Mode, an optional security setting to protect against prompt injection attacks. Lockdown Mode limits some features in ChatGPT, such as image generation and file downloads, but allows image uploads and conversation sharing. It is designed for users handling sensitive data and is available to all personal accounts, including those on the free tier. Enabling Lockdown Mode prevents attackers from extracting sensitive data from accounts by limiting network requests that could be exploited.

Amazon Bedrock releases a new console experience optimized for Anthropic and OpenAI-compatible APIs. The console supports GPT, Claude, and open-weight models via bedrock-mantle endpoint with OpenAI Responses, Chat Completions, and Anthropic Messages APIs. Users can compare up to three models side by side and view token usage, inference requests, and pricing per model. The project dashboard shows real-time inference metrics and model distribution by date and usage.