OpenAI disclosed a self-replicating prompt injection that spread across simulated agent workflows, copying itself through outputs much like a computer worm. The OpenAI Alignment report shows injected instructions propagating through email replies, filesystem artifacts, code…
Google added Live Avatar to Gemini 3.8 Live on September 24, giving its conversational model a near-real-time on-screen persona and making it available through Gemini Enterprise. According to the Google documentation, the model can return synchronized speech and avatar video at…
DeusData has released an open-source MCP server for code repositories that builds a persistent graph AI coding agents can query across 162 programming languages. The codebase-memory-mcp project runs indexing and querying locally, without an embedded LLM or a required API key…
Alibaba previewed the Qwen 4 lineup at its Apsara conference on September 22, but stopped short of launching a callable product. The Qwen 4 overview describes four models still in training: Qwen 4 Max as the flagship, Qwen 4 Flash for throughput and latency, Qwen 4 Plus as the…
Meta expanded Muse into email, commerce and smart glasses at Meta Connect 2026, while unveiling a pocket-sized Muse Charm and new wearable hardware in its Connect announcement. Muse for Mac can now operate permitted apps and continue queued work unattended. Meta also listed Shop…
OpenAI has released MentalHealthBench, a public benchmark for testing how models handle mental health conversations across routine stress, high-acuity situations, and emergencies. The dataset and method are outlined in the OpenAI announcement. The release includes 1,215…
Cursor changed its agent harness to reduce user token costs by 7% without lowering agent quality, the company says. Cursor’s breakdown attributes the savings to a system prompt trimmed by roughly 66%, dynamic tool loading that cut static-context description tokens by 60%, and…
Google launched Gemini 3.8 Flash TTS and Gemini 3.8 Flash-Lite TTS on September 23, adding a higher-expression speech model alongside a lower-cost option for high-volume audio work. In the Google announcement, the company says Flash is aimed at more directed voice production…
NVIDIA released Nemotron 3 Diarization, an open-weight audio model for tracking speakers in live and recorded conversations. The roughly 100-million-parameter checkpoint labels up to eight speakers, including overlapping speech, according to the launch post. The model returns…
Anthropic and OpenEvidence said on Sept. 22 they are expanding a free clinical decision-support service for healthcare workers across about 100 low- and middle-income countries. The Reuters report says the rollout includes countries such as Uganda, Angola, Sudan, Haiti, and…
OpenAI launched GPT-6 Sol and GPT-6 Luna on September 22 as lower-cost additions to its GPT-6 model line. According to The New Stack, Sol is priced at $2 per million input tokens and $10 per million output tokens. Luna is listed at $0.10 for input and $0.50 for output per…
Anthropic announced Claude Opus 5.5, the first model in its Claude 5.5 family, and said it is available through Anthropic, AWS, Google Cloud, and Microsoft Azure. In the API, it appears as claude-opus-5-5. In the Anthropic announcement, the company lists standard pricing at $4…