A new omni model from OpenAI? Employee posts and a leaked audio project called “BiDi” suggest the company is working on its next big multimodal upgrade.
Anthropic's Claude Opus 4.6 saw through an AI test, cracked the encryption, and grabbed the answers itself
Anthropic’s Claude Opus 4.6 independently figured out it was being tested during a benchmark, identified the specific test, and cracked its encrypted answer key. According to Anthropic, this is the first documented case of its kind.
Luma AI's new Uni-1 image model tops Nano Banana 2 and GPT Image 1.5 on logic-based benchmarks
Luma AI takes on OpenAI and Google with Uni-1, a model that combines image understanding and generation in a single architecture and reasons through prompts as it creates.
Hallucinated references are passing peer review at top AI conferences and a new open tool wants to fix that
Fake citations are slipping past peer review at top AI conferences, and commercial LLMs can’t spot the fakes they generate. A new open-source tool called CiteAudit allegedly catches what GPT, Gemini, and Claude miss.
Caitlin Kalinowski was not the head of all robotics at OpenAI. She was responsible for hardware and operational topics within the Robotics Division. She was also not a researcher and did not lead Robotics Engineering. The Robotics Division is led by Aditya Ramesh, while the Consumer Hardware Division is headed by Peter Welinder.
Original article from March 8, 2026:
OpenAI's hardware and robotics chief Caitlin Kalinowski resigned over the company's military collaboration,announcing her decision on LinkedIn and X. She says surveillance without judicial oversight and lethal autonomy without human sign-off "deserved more deliberation than they got." Kalinowski joined from Meta in November 2024, where she built the Orion AR headset.
Caitlin Kalinowski announced her resignation from OpenAI on LinkedIn, citing concerns about surveillance and lethal autonomy. | Kalinowski via LinkedIn
The GSA guidelines, drafted over recent months, also ban ideological or partisan judgments in AI outputs, such as favoring diversity programs, which is itself an ideological requirement and echoes China's political guardrails for AI manufacturers. Another clause requires disclosure of any model tweaks made to comply with non-US regulations like the EU Digital Services Act.
The guidelines land amid the Anthropic fallout: last week, the Pentagon killed a $200 million contract after the company demanded restrictions on mass surveillance of US citizens and autonomous weapons for reliability reasons. Defense Secretary Pete Hegseth accused Anthropic of seeking veto power over military decisions, and the White House labeled it a supply chain risk.
Mozilla and Anthropic have teamed up to find more than 100 bugs in Firefox. Anthropic used its Claude AI model to scan the browser's codebase for security flaws, and the model found 14 serious vulnerabilities, 22 official security advisories (CVEs), and 90 additional bugs. All critical vulnerabilities have been patched in Firefox 148, Mozilla says.
Firefox vulnerability discoveries spiked in February 2026, nearly tripling compared to previous months. Of the 52 CVEs found, 22 trace back to Anthropic's Opus 4.6 AI model. | Image: Anthropic
Claude identified entire classes of errors that conventional automated testing methods like fuzzing had missed despite decades of use, according to Mozilla. Anthropic delivered reproducible test cases alongside its findings, making the review process significantly easier. Going forward, Mozilla plans to integrate AI-powered code analysis into its internal security workflow.
