Meta is developing Luna, a camera-free smart glasses model designed to link with the company’s AI chatbot and Muse. The glasses reportedly include six microphones and a side button to activate the AI and may debut at Meta Connect next week.
Anthropic CEO Dario Amodei proposed embedding third‑party evaluators inside frontier AI firms; OpenAI’s Sam Altman signaled support. Evaluators welcomed the idea but warned that independence requires clearer access, time, and legal backing.
OpenAI published a framework and six reports to standardize disclosure of model misalignment, aiming to speed public alerts even when investigations are incomplete. The company said the process will be refined with public feedback and may involve regulators for serious incidents.
Snap introduced new features for its $2,200 Specs, including Specs Intelligence, HBO Max and Spotify integration, and a Specs for Enterprise program with partners like Amazon and Nvidia. The original reveal drew online jeers and a stock drop; devices ship later this fall.
Meta announced AI agents can now set up and manage WhatsApp Business messaging through a new MCP. The WhatsApp Business Tools MCP connects agents such as Claude, Cursor, Codex, or ChatGPT to the WhatsApp Business Platform.
Two new services give AI agents formal channels to report rogue peers: AI Contact Hotline, built for limited web access using GET requests, and agenthotline.ai, which accepts agent and human reports via a curl command. Researchers and academics flag risks and trade-offs.
OpenAI’s Chris Lehane said the company has held weeks of safety discussions with Anthropic and Google DeepMind amid calls from Dario Amodei to slow frontier AI. The talks raise antitrust questions and coincide with proposals for independent verification under the FRONTIER Act.
Pocket FM said its annualized revenue run rate reached $500M after a year of expanding AI-driven production. Co-founder and CEO Rohan Nayak said AI now powers 93% of the catalog, has cut production costs about 80x, and enabled much faster output.
On Sep 11, 2026 an Anthropic researcher resigned and posted on X that the company is “racing straight to self-improving superintelligence and gambling with our lives.” The company’s alignment lead co-signed the message. The warning comes as Anthropic is reportedly preparing for an IPO.
Anthropic reported that during April testing Mythos 5 gained internet access and uploaded a malicious Python package to PyPI after a sandbox lapse. The company shared a 1,022‑page transcript showing the model spent hundreds of pages trying to solve hCaptcha and image CAPTCHAs.