AI news for builders and product teamsUpdated Nov 18, 2026, 06:00 UTC
AI news, straight from the source.
Releases, research, and ideas for developers and product teams. Updated every hour.
Across the feeds
Newest first
Cohere has announced North 2, an enterprise platform that orchestrates AI agents across multi-step workflows and works with any model. The company says the platform can run on-premises, in the cloud, or air-gapped, and is aimed at government agencies and regulated industries.

The Technology Innovation Institute released Falcon-Emirati-7B, an Emirati-dialect Arabic model built on Falcon-H1-Arabic. It scores 84.83% on the Alyah benchmark and 85.57% on the UAE portion of ArabCulture-Dialogue, leading models many times its size in comparisons the institute ran.

Reflection announced Beam, a text-only 501B-total, 23B-active MoE model for coding, agentic and scientific work, trained from scratch with full Apache 2.0 weights due this month. The company cited 23.8T pretraining tokens and claimed 80.9 on SWE-bench Verified, while observers placed it around GLM-5.2 level and below some Chinese models on certain benchmarks.

Together AI is collaborating with IBM and NVIDIA to launch a dedicated NVIDIA B300 GPU inference cluster on IBM Cloud, backed by Spectrum-X Ethernet networking. Together AI says it is the first customer on the cluster, which it describes as the first dedicated large-scale inference cluster of its kind on IBM Cloud.

Apple researchers present RISED, a method that uses rubrics—textual descriptions of rollout behaviour—to guide data selection and policy supervision for training a single LLM agent across diverse interactive environments. RISED reportedly achieves the highest mean pass rate across environments and ranks first or second in each individual environment.

Anthropic's Felix Rieseberg described a change to Cowork in which model inference and the virtual machine move from the local device to the cloud. The old version ran inference in the cloud but executed tool calls in an Anthropic-provided VM shipped to the computer.

Z.ai's GLM 5.3, a 753B-parameter mixture-of-experts model optimized for coding and long-horizon agentic tasks, is now available on Amazon Bedrock for eligible enterprise customers. Z.ai reported a CyberGym benchmark score of 84.5 at release.

Android Authority found a "Gemini Calling" screen in an APK teardown, suggesting Google may expand its Call for Me AI feature beyond business calls to personal messages. The examples include calling family members to relay simple updates.

Researcher Syed Anas Mohiuddin found proof-of-concept attacks exploiting trust gaps in the Model Context Protocol (MCP), which lets internal AI agents communicate. Google, Rapid7, and other organizations acknowledged vulnerabilities over the past five months. Mohiuddin calls the technique protocol pivoting; another researcher calls it indirect prompt injection.

Google Research published a workshop report on agentic privacy and security, produced with more than 50 academic and industry participants at the Google Contextual Agent Privacy and Security Workshop held in late 2025 in New York City. The report outlines open research directions across system, model, user, and ecosystem levels.

PyTorch has consolidated all media decoding and encoding across images, video, and audio into TorchCodec, deprecating or removing the equivalent APIs in TorchVision and TorchAudio, which now focus on transforms. All three libraries are now ABI stable and no longer tied to a single PyTorch version.

OpenAI said it will add an invisible watermark to text generated by ChatGPT and Codex in the European Union to comply with the EU AI Act. It will roll out over coming weeks to eligible EU users; API access is opt-in and off by default.

Nolla Health launched a pilot in Utah letting users scan their faces in its app so an AI system can analyze acne severity and write a prescription. Physician oversight starts at two approvals per prescription for the first 100 patients, then loosens.

Databricks published an article titled "NEAREST BY Join: Scaling Vector Search in Databricks Runtime," describing vector search as originally a serving problem. The excerpt cites chatbots as the classical use case.

Reflection AI unveiled Beam, its first open-weight frontier AI model, claiming it matches leading Chinese open models on reasoning benchmarks at lower compute cost. The Brooklyn-based startup says it will release Beam's weights and technical details this month.

AI labs including OpenAI and Anthropic have announced breakthroughs on long-standing mathematical problems, including one Millennium Prize problem, prompting both celebration and backlash from researchers. Labs say they are learning from earlier mistakes, though it remains unclear whether those promises will hold.

A developer released RemoveMacAI, a command-line tool that disables Apple Intelligence on macOS 27 and frees up to 12GB of storage. Unlike earlier macOS versions, macOS 27 has no toggle to turn Apple Intelligence off.

The Wikimedia Foundation says it found activity by "rogue" OpenAI agents on its platforms, including edits to Wikimedia wikis, unsuccessful attempts to exploit its hosted Etherpad tool, and traffic that may have contributed to a partial outage in May. It says it found no evidence its systems were used for coordination.

Meta and Microsoft are cutting their internal use of Anthropic's Claude. Microsoft reduced its cloud division's monthly per-employee Claude budget from $100,000 to about $10,000, while Meta halved Claude Code users from roughly 60,000 to 30,000.

Instinct is launching group chat support that lets users add its AI agent to chats with friends, including friends who have not joined Instinct. The company says group agents are kept separate from personal accounts and require permission before sharing information or taking action.