Artificial intelligence news.
Straight from the original announcements.

MAIEI previews SAIER Volume 8: oversight, sovereignty and surveillance in 2026

October 13, 2026, 20:24Source: Montreal AI Ethics Institute
MAIEI previews SAIER Volume 8: oversight, sovereignty and surveillance in 2026
Image: Montreal AI Ethics Institute

The Montreal AI Ethics Institute has previewed the State of AI Ethics Report (SAIER) Volume 8, Power, Fracture, Resistance, which is due to publish on November 10. According to the institute, contributor drafts point to six themes from 2026, from the spread of the promise that "humans decide" to the places where the costs of AI infrastructure and surveillance land.

Key points
  • SAIER Volume 8, Power, Fracture, Resistance, is due to publish on November 10, according to the Montreal AI Ethics Institute.
  • Contributors say the "humans decide" promise, seen in a Brazilian guide and a UN statement, risks becoming compliance theatre without protecting the capacity to exercise oversight.
  • A Pacific Islands Forum communiqué reaffirmed 38 times, endorsed 20 and tasked once, which a contributor uses to contrast sovereignty talk with sovereignty terms.
  • Alberta accounts for 92 percent of Canada's planned data centre capacity by megawatt, and at least three projects were found not to need a provincial environmental impact assessment.
  • UNESCO data cited in the brief show 75 percent of surveyed states with an AI strategy reference ethics, but only 16 percent report dedicated redress or liability frameworks.

A report built around what AI asks in return

The preview appears in The AI Ethics Brief #201, the institute's bi-weekly publication. Its premise, first announced in May, is that every AI promise asks for something in return. People give up friction and receive efficiency, but increasingly they also give up decisions, deliberation and accountability, often before anyone has agreed the trade is acceptable.

First drafts arrived on September 18, and final pieces were due a week after the brief was published. Contributors were told to write no evergreen analysis, only what happened in 2026 and what it signals for 2027. The same edition also includes an AI Policy Corner, produced with GRAIL at Purdue, which says Washington DC frames AI as competition with other states and cooperation with private firms. The institute says that raises questions about confidential procurement, corporate interests and who sets the rules for military AI.

"Humans decide" becomes a formula

The first theme is a promise repeated across the drafts: a human must decide. The brief points to two statements a month apart. On June 25, Brazil's Ministry of Management and Innovation in Public Services published a public-sector guide whose checklist reads, in a contributor's translation, "AI suggests, but the human decides and assumes legal responsibility". In July, at the first Global Dialogue on AI Governance in Geneva, the UN Secretary-General told more than 170 countries that machines "can inform, but humans must decide, and answer".

Contributors test whether those sentences hold. One argues the Brazilian guide names human oversight as the central safeguard but leaves unprotected the capacities people need to exercise it, so that "oversight becomes compliance theatre: it attests to what it does not ensure". Another proposes a participatory fundamental rights impact assessment under Article 27 of the EU AI Act, with dissent documented and an anonymous final vote on deployment. A piece on home care in Britain describes a human who still signs a care decision, while the evidence now comes from an in-home sensor or a voice-activated robot. Councils must give written reasons, but nothing requires them to say which evidence came from a device or what it could not capture.

Sovereignty talk, and access that can be switched off

The second theme is the gap between sovereignty rhetoric and binding terms. Pacific participation in Geneva rested on a sponsorship scheme in which, as the organizers put it, "the number of candidates selected depended on funds available." Later, Pacific Islands Forum leaders endorsed the Pacific Digital and Artificial Intelligence Technology Principles in Koror, with a framework due in 2027. Tess Newton Cain and Stefan Armbruster counted the communiqué's verbs: reaffirmed 38 times, endorsed 20, tasked once. "Sovereignty talk is now abundant," the contributor writes. "Sovereignty terms are not".

From the supply side, another contributor notes that ninety percent of the computing power needed for frontier AI is located in the United States and China, and that most national programmes follow what they call "hybrid sovereignty", with governments co-developing infrastructure with private partners.

The brief also describes access as something that can be withdrawn. In June 2026, it says, Anthropic temporarily restricted foreign access to two of its leading models, Mythos and Fable, to comply with a US government directive, leaving institutions with no role in the dispute without access. One contributor argues that "reliability of accessibility" is now a dimension of value in model choice. The same contributor cites 2026 UNESCO implementation data: among surveyed member states with an AI strategy, 75 percent reference ethics and human rights, but only 16 percent report dedicated redress or liability frameworks for AI-related harm.

Community-held governance and local costs

On participation, contributors point to the PAVE Case Book, launched alongside the Geneva Dialogue, which catalogues over 80 participatory processes worldwide. They also look to Indigenous and African models. Te Mana Raraunga and Maiam nayri Wingara each moved from principles to a binding agreement with the state to an operational body. The Nwulite Obodo Open Data License keeps African-language datasets royalty-free for researchers in lower-income countries while imposing benefit-sharing obligations on users in high-income ones. The Esethu Framework routes licensing fees from non-African commercial users back into expanding local datasets.

The brief then turns to who pays. Alberta accounts for 92 percent of Canada's planned data centre capacity by megawatt. At least three Alberta projects, including Wonder Valley, have been determined not to require a provincial environmental impact assessment. The brief stresses that no existing requirement was lifted: data centres are simply not classified as activities that automatically trigger one. In Hamilton, Ontario, a proposed data centre drew over a thousand public comments, and the Planning Committee unanimously backed a temporary moratorium, but the motion failed at council.

Surveillance as care and what to expect

The sixth theme is surveillance presented as care, safety and convenience. The brief lists UK age verification expanding, Canada's C-22 lawful access bill passing the House of Commons in June, and the US KIDS Act clearing the House the same month, all framed as child protection. The European Parliament rejected the Council's position on Chat Control, but the regime was extended anyway. One contributor argues these measures reverse the burden of proof, requiring users to prove themselves by default. Another piece follows Oxevision, a camera system that tracks a patient's movement, breathing and pulse inside NHS mental-health bedrooms without staff entering the room.

For policymakers, procurers and communities, the preview suggests the 2026 debate is shifting from stated principles to enforceable terms: who can inspect evidence, who holds access, and who bears local costs. The full volume is due on November 10.

Related news

UN launches toolbox to keep people accountable for AI in smart cities
BusinessNovember 6, 2026, 14:43

UN launches toolbox to keep people accountable for AI in smart cities

The United Nations has launched its first system-wide toolbox for using artificial intelligence in smart cities while keeping people accountable for public decisions.…

UNESCO-led AI supervisors' network grows to 42 authorities, shifts to practice
BusinessNovember 2, 2026

UNESCO-led AI supervisors' network grows to 42 authorities, shifts to practice

The Global Network of AI Supervisory Authorities (GNAIS), led by UNESCO, has expanded to 42 AI supervisory authorities in its first year, according to UNESCO. At the 4th…

AI Ethics Brief: Consumer Agents Shift Decisions Away From Users
BusinessOctober 27, 2026, 22:11

AI Ethics Brief: Consumer Agents Shift Decisions Away From Users

The Montreal AI Ethics Institute argues in its latest brief that consumer AI agents such as Meta's Muse and OpenAI's dots move key decisions away from the people they…