AI news5
OpenAI begins rolling out Dots, always-on agents with their own cloud computers
OpenAI introduced Dots on 29 September: agents powered by GPT-6 Astra that continue working between conversations. Users give them goals, connect apps and set their autonomy. Access is rolling out to eligible Pro and Business Premium users in supported markets; Enterprise beta access is off by default.
help.openai.com ↗Why it matters
This brings ongoing delegated work into ChatGPT, a new OpenAI release distinct from earlier Microsoft and Perplexity always-on agents.
Anthropic releases Sonnet 5.5, claiming faster work at lower cost per task
Anthropic released Claude Sonnet 5.5 on 28 September. It reports over 30% faster output and up to 30% lower cost per task through reduced token use. API rates remain $2 per million input tokens and $10 per million output tokens.
anthropic.com ↗Why it matters
The Sonnet release anticipated in last week's Opus announcement has now arrived, giving teams a new option to evaluate for everyday coding and document work.
Google announces Gemini 4 Argon, initially restricted to trusted cyber defenders
Google announced Gemini 4 Argon on 30 September, targeting complex coding, professional work and cybersecurity tasks. Initial access is through its Fairwind programme for selected cyber defenders. Wider developer, enterprise and consumer access will follow after further testing; introductory API prices are planned at $2 per million input tokens and $10 per million output tokens.
blog.google ↗Why it matters
Access and safeguards are becoming part of the product rollout, so a frontier-model announcement does not immediately create a tool every team can use.
Nvidia releases OpenShell software for enforcing AI agent boundaries
Nvidia announced its Open Agent Safety Platform on 28 September, with OpenShell software available through developer resources and GitHub. OpenShell enforces runtime policies, while the Sentry reference design adds a separate watchdog on BlueField-4 processors. Nvidia says Sentry can stop agents crossing their boundaries within milliseconds.
nvidianews.nvidia.com ↗Why it matters
Controls outside the agent provide another layer of enforcement when instructions inside a model prove insufficient.
Nasdaq launches a governed AI agent environment inside Calypso
Nasdaq announced new Calypso capabilities on 29 September, starting with a natural-language assistant for querying platform data and documentation. The environment uses Amazon Bedrock and MCP connections, with operational boundaries, oversight and sandboxing. Additional agent workers are planned, with clients working through their own governance requirements before activation.
press.aboutamazon.com ↗Why it matters
This is a new Calypso product launch, distinct from prior Nasdaq stock-market coverage, and shows how agents are entering software used for regulated operations.
AI in the nonprofit sector3
New AI Equity Project findings expose a nonprofit governance gap
Namaste Data's 2026 AI Equity Project report was scheduled for release on 29 September and now has a download link. Its 880 nonprofit respondents represent organisations in the US and Canada. Nearly 70% reported some organisational AI use, while only 12% said it was guided by organisation-wide practices or expectations.
namastedata.org ↗Why it matters
Teams need shared rules and accountability alongside tool training; widespread individual experimentation does not establish organisational readiness.
Blackbaud unveils Lantern fundraising model, with first features planned for 2027
Blackbaud announced Lantern on 29 September, developed with Databricks and trained on synthetic fundraising scenarios based on its Social Impact Signal Graph. Hesed House and YMCA of the North are among customers contributing input. It is planned to power selected Raiser's Edge NXT and Agents for Good features in 2027.
blackbaud.com ↗Why it matters
Sector-specific models may improve fundraising workflows, but this announcement provides a roadmap to assess rather than evidence of results nonprofits can reproduce today.
Free fundraising workflow session runs tomorrow at 4:00 a.m. AEST
OpenAI Academy's Alex Nawar will demonstrate reusable fundraising workflows at CauseVox's free online Impact Overflow summit. The session uses a year-end campaign to show source review, planning, reusable Skills and recurring updates with human approval points. It runs Friday 2 October, 4:00 to 4:45 a.m. AEST, equivalent to Thursday 1 October, 2:00 to 2:45 p.m. US Eastern time.
academy.openai.com ↗Why it matters
A concrete campaign example can help fundraising teams identify repeatable work while keeping donor relationships and judgment with people.