AI news5
Google releases Gemini 3.8 Flash TTS models that can clone a voice from 30 seconds of audio
Google launched Gemini 3.8 Flash TTS and Gemini 3.8 Flash-Lite TTS, rolling out in the Gemini API and AI Studio. Users can design new voices from a text prompt or pick from more than 2,000 presets, and Flash TTS can recreate a voice from a 30-second sample after a consent check. Google says Flash TTS topped Hume AI's Voice Design Benchmark at 71.4 and that both models placed first and second on Hume's Overall Quality Index, with Figma, HeyGen, Wondercraft and Ollang among the first integrators.
thenextweb.com ↗Why it matters
Voice cloning from a half-minute sample with a consent gate pushes synthetic voice into ordinary product teams, and consent verification is what will decide whether this lands as a tool or a liability.
Xiaomi releases the MiMo-V2.6 trillion-parameter model family
Xiaomi released its MiMo-V2.6 family, including MiMo-V2.6-Pro, MiMo-V2.6-Flash and an ultra-speed variant for latency-sensitive work. Xiaomi describes the Pro model as an omni-modal trillion-parameter reasoning system built for long-horizon tasks, research, cybersecurity and complex professional workloads, while Flash targets lower-cost, high-volume use. It is also releasing model components and tools to support research into reinforcement learning and agents.
techstartups.com ↗Why it matters
A trillion-parameter open model from a consumer-device maker extends the cost pressure that is already forcing US labs to halve API prices.
Nvidia launches Isaac ROS 5.0 to bring AI agents deeper into robotics development
Nvidia released Isaac ROS 5.0, the latest version of its GPU-accelerated software stack for developers building on the Robot Operating System, announced at ROSCon in Toronto. The release adds AI-agent skills so coding assistants can help developers configure environments, migrate software and work with Nvidia's robotics tools, adds support for ROS 2 Lyrical, and expands deployment across the Jetson edge-computing platform.
techstartups.com ↗Why it matters
Putting agents inside the robotics toolchain is how Nvidia courts robot developers before they pick a compute vendor, the same lock-in pattern it ran in AI training.
Sam Altman to brief the UN Security Council on AI safety
OpenAI CEO Sam Altman is due to brief the UN Security Council in person, with Anthropic's Dario Amodei joining by video alongside Yoshua Bengio and Hugging Face's Clement Delangue, chaired by France. It is the Council's first meeting focused on the safety risks of increasingly capable AI, including loss of human control. It follows OpenAI's 21 September proposal for US-led global standards on recursive self-improvement, and comes the same week Treasury Secretary Scott Bessent said the US and China agreed to open an AI dialogue.
thenextweb.com ↗Why it matters
The UNSC taking up frontier AI for the first time puts loss of control onto the same agenda as nuclear and security risks, even as the US President publicly dismisses a global rulebook.
Qualcomm unveils Snapdragon 8 Elite Gen 6 on TSMC's 2-nanometer process
Qualcomm announced the Snapdragon 8 Elite Gen 6 and 8 Elite Extreme Gen 6, each built on TSMC's 2nm process with two 5GHz prime cores and six 4GHz performance cores. The chips push more AI inference onto phones rather than servers, and add another anchor customer to TSMC's N2 ramp. The launch came alongside Cognate agreeing to buy Intel's RealSense 3D vision business for about $600 million, and Alphabet's Intrinsic open-sourcing Intrinsic Core under Apache 2.0.
marketanalysis.com ↗Why it matters
On-device inference at 2nm is what makes local AI features viable without a data-centre round trip, which matters most for privacy-sensitive and offline use cases.
AI in the nonprofit sector4
Young Futures opens The Unwritten Challenge, at least $1M for 10+ US youth organisations
Young Futures opened applications for The Unwritten Challenge, supporting nonprofits helping young people aged 10 to 19 learn deeply, think critically and make healthy decisions in an AI-shaped world. It will distribute a minimum of $1 million across at least 10 organisations, with awards of $50,000 to $100,000, and applications close 16 October 2026. It is the first of two US funding challenges under Groundwork, Young Futures' $15 million three-year US and UK initiative, with an applicant information session on 29 September.
financialcontent.com ↗Why it matters
A named, small-org-friendly grant with a youth-in-design requirement is a concrete entry point for the youth-serving organisations being asked what their AI response is.
MTN Group Foundation and the Gates Foundation launch a $25M AI maternal health push in Nigeria
The MTN Group Foundation and the Gates Foundation announced the Nigeria Maternal Health Multiplier, a four-year partnership beginning with an initial $25 million between 2026 and 2030, half direct and half in kind. The initiative will deploy AI-enabled tools to improve maternal health outcomes for 500,000 women in Nigeria by 2030, complementing the Federal Ministry of Health's maternal and neonatal mortality reduction programme, and is intended to catalyse further funding and expansion into other African markets. It was announced at Semafor's Next 3 Billion event during the UN General Assembly.
healthwise.punchng.com ↗Why it matters
A telecoms operator and a major funder pairing local distribution with AI health tools is a replicable model for reaching frontline patients outside traditional health systems.
Verizon commits $70M to AI skills, pairing labs' training with community nonprofits
Verizon announced Verizon AI Skills for America, a $70 million investment combining its existing $20 million reskilling fund with $50 million in new funding. The programme pairs training from IBM, Google, Anthropic, Microsoft, Coursera and OpenAI with local delivery through mission-based organisations including the Local Initiatives Support Corporation, the National Association for Community College Entrepreneurship and Goodwill Industries International, with core wrap-around services in select regional markets. It targets job seekers, displaced workers, educators and small businesses.
verizon.com ↗Why it matters
The programme's stated logic, that access to content alone does not work without local coaching, is the same conclusion nonprofit AI enablement projects keep reaching.
Coefficient Giving opens a $200M AI safety fund and a $50M fund-seeding call, EOIs close 2 October
Coefficient Giving, formerly Open Philanthropy, is running two open calls. Project Tailwind is a roughly $200 million rolling call for founders to launch new AI safety initiatives, funding pre-seed grants of $200,000 to $2 million with no legal entity required, and publishing a list of initiatives it would fund such as rogue-incident detection, verification technology and an independent AI auditing body. Launchpad Funds is a $50 million call to seed or scale new funds in global health and wellbeing, with expressions of interest closing 2 October 2026.
grantedai.com ↗Why it matters
Both calls recruit founders rather than review proposals, which is philanthropy acting as a field-builder and the clearest current route to starting a new AI safety organisation.