AI news6
DHS publishes notice of a DeepSeek removal order dated 24 July
A DHS notice filed on 19 October describes an existing order excluding DeepSeek products from federal civilian procurement and requiring their removal from agency IT. The scope includes third-party products developed using DeepSeek implementations. The scanned order is dated 24 July; Federal Register publication is scheduled for 20 October.
public-inspection.federalregister.gov ↗Why it matters
This new public filing documents procurement and supply-chain restrictions, a separate development from DeepSeek's recently covered model and API releases.
Anthropic reports junior technical hires falling from 27% to 9%
Anthropic's 16 October workforce study reports that juniors' share of technical hires fell from 27% to 9%, while technical hiring still grew in absolute numbers. By August, employee-directed agents logged six times their users' working hours. Parallel agent runtime is not equivalent to six human workers.
anthropic.com ↗Why it matters
This new internal workforce study adds hiring-mix evidence beyond earlier Anthropic model announcements; one AI lab's experience does not establish economy-wide causation.
XRlabs raises US$8.4 million for its surgical AI platform
XRlabs announced a US$8.4 million seed round on 19 October, led by Playground Global with Eka Ventures participating. The company says its surgical AI platform has been used in more than 100 cases across three continents. Funding will support engineering, clinical evidence and regulatory work.
financialcontent.com ↗Why it matters
The round backs AI integrated with existing operating-room equipment; the deployment figures are company-reported and do not establish regulatory approval or clinical benefit.
OpenCode discloses a breach affecting 1.4% of Console users
OpenCode's 17 October statement says a security breach linked to a large-scale distillation attack affected 1.4% of Console users. It says the affected subset had external model providers connected and was notified by email. The company described remediation with Anthropic, OpenAI and Google as ongoing.
x.com ↗Why it matters
This newly dated incident concerns linked-provider exposure, rather than the older generic OpenCode mentions found in the archive; the initial statement does not prove remediation is complete.
Microsoft outlines three layers for supervising internal AI agents
In a 19 October security article, Microsoft describes applying Project Perception to agents inside an organisation. Its three layers combine identity and permissions, changing risk signals, and Guardian Agents that supervise actions. Project Perception was announced in July; this article sets out its internal-agent security approach.
blogs.microsoft.com ↗Why it matters
The new internal-agent application goes beyond Microsoft's previously covered weather and infrastructure work, putting continuous oversight alongside agent deployment.
MyLens introduces private, interactive one-to-one presentations
MyLens's creator introduced its new presentation workflow in a 16 October Product Hunt post. Users supply source material, then each viewer gets a private AI presentation that can answer questions and generate additional slides during the session. Creators can review individual session questions and follow-up needs.
producthunt.com ↗Why it matters
It offers a concrete format for reusable onboarding and teaching, although the maker's feature claims are not evidence of better learning outcomes.
AI in the nonprofit sector4
Ersilia's US$10,000 African research incubator closes at 8:59 AM AEDT
Ersilia and the H3D Foundation offer US$10,000 per selected project for a 24-week AI drug-discovery programme running February to July 2027. Eligible teams of up to four work at academic or nonprofit institutions in the WHO African Region on existing small-molecule infectious-disease projects. Applications close at 23:59 CEST on 19 October, equivalent to 8:59 AM AEDT today, 20 October.
incubator.ersilia.io ↗Why it matters
The verified deadline makes this a final-day funding opportunity; applicants also need a budget and signed participation and data-sharing agreement.
Superbloom offers US$25,000 for community-led AI feedback workshops
Superbloom's 6 October call offers each selected New York community partner a US$25,000 stipend to host two bilingual English/Spanish workshops. It seeks at least three New York City organisations and one upstate partner with strong Latino community relationships. The page lists 19 October as the deadline but gives no closing hour.
superbloom.design ↗Why it matters
The programme funds community input into AI transparency standards; the unspecified closing time means the date alone does not confirm applications remain open.
Anthropic proposes putting 5% of its shares into a public-benefit fund
Anthropic's 18 October announcement commits 5% of its shares to a new Public Benefit Fund, subject to shareholder approval. Its board could add two percentage points for each 50% valuation increase above a set baseline. The wholly owned LLC may support nonprofits or direct distributions, but the announcement provides no application round or deadline.
anthropic.com ↗Why it matters
This new equity-backed giving structure is distinct from earlier Anthropic grants; proposed capacity should not be counted as grants already awarded.
OpenAI offers eligible research labs up to 30 free Rosalind seats
OpenAI's 15 October Rosalind Academic Plan offers eligible US university labs up to 30 Business Standard seats at no cost for up to 12 months. Access to the life-sciences model requires trusted-access approval, and API access is excluded. Labs at universities with an existing Enterprise agreement must use their institution's access process.
openai.com ↗Why it matters
This newly announced academic-access programme lowers a research resource cost, while its US-only eligibility and approval requirements limit who can use it.