OpenAI made ChatGPT Lockdown Mode available to all logged-in users and added moderation scores to API generation requests on June 4. The changes move prompt-injection and data-exfiltration defenses from policy language into product controls.
Personal AI is shifting from single-session answers to durable context. OpenAI says the new ChatGPT memory starts with US Plus and Pro users and uses a more efficient architecture that reduced compute needs by about 5x.
OpenAI’s June 3 blueprint turns state frontier-AI bills into a proposed federal template. The plan centers on CAISI, independent audits, severe-risk evaluations, incident reporting, model-weight security, and a broader government resilience strategy.
Codex is expanding from developer assistant into role-specific work surfaces. OpenAI says the new plugins bundle 62 apps and 110 skills, while Sites adds shared interactive pages for Business and Enterprise teams.
Life-science AI is moving from literature help toward executable research workflows. OpenAI says GPT-Rosalind reached 27.5% on MedChemBench, 21.6% on GeneBench, and 63.2% on LabWorkBench.
OpenAI frontier models and Codex are now generally available on Amazon Bedrock. The post drew more than 1.2M views and points to a broader AWS path for enterprise AI, including future Daybreak security capabilities.
OpenAI is moving its robotics work from research framing toward hiring for hardware, operations, systems, and ML execution. Sam Altman’s post drew more than 2.16 million views and 11,400 likes, making embodied AI the clearest signal in this 48-hour X window.
OpenAI is moving frontier AI deeper into biodefense, not only biomedical discovery. The post says Rosalind Biodefense and GPT-Rosalind access will support selected U.S. government and allied public-health missions.
OpenAI moved Codex computer use onto Windows, changing where coding agents can act. The May 29 post passed 950,000 views and says ChatGPT mobile can start, review, and redirect work running on a Windows machine.
OpenAI is moving its election playbook from general guidance to live data and provenance checks. For the US and Brazil, ChatGPT will point to AP vote counts, while a public tool will test OpenAI-origin SynthID watermarks and C2PA metadata.
OpenAI is integrating both C2PA Content Credentials and Google SynthID invisible watermarks into images generated by ChatGPT, Codex, and the OpenAI API. The dual-layer approach provides mutual redundancy — C2PA carries detailed metadata while SynthID survives image transformations. A public Verify tool is now available in preview for checking AI image provenance.
OpenAI says a general-purpose reasoning model found a construction disproving the conjectured upper bound in Erdős's planar unit-distance problem. Mathematicians reviewed the proof, but the ML community raises questions about methodological transparency.