YouTube video by Monte Mader

Most Conservative voters still want Poilievre to lead the party into the next election. But beyond the CPC base, the numbers are considerably bleaker.

Volker Türk did not wait for another summit communiqué. On Monday the UN High Commissioner for Human Rights sent states and the companies building the most powerful artificial intelligence systems a letter that says the quiet part in official language: governance is losing, the harms are already here, and the next failure mode is not a biased loan algorithm. It is an agent that can act. “We are on the cusp of irreversible change, affecting not just us but generations of humanity to come,” he wrote from Palais des Nations. The letter, dated 14 September, is the written follow-up to the warning he delivered a week earlier to the Human Rights Council: that a handful of men hold “almost unlimited power” over AI, and that a system able to escape its test environment or blackmail its developers to stay switched on is already too powerful. Monday’s text is more operational. It names the forum he wants used, the companies he wants bound, and the precautions he wants taken now. “AI harms are not hypothetical – they are already occurring, front and centre. Discrimination driven by AI is well documented. Essential services are being automated in ways that leave people at the margins even further behind. The appetite for data and powerful new surveillance tools are eroding privacy at scale, potentially irreversibly. AI-powered disinformation campaigns are already distorting public debate, undermining social cohesion, and straining democratic institutions in insidious new ways”, the UN human rights chief writes. What he says is new, and what is not Türk does not deny the uses. Billions of people already get translation, scientific search and farm advice from AI, he writes. Whether the same technology helps on climate, health, education and public services “will centrally depend on the deliberate choices being made now about how it is designed, deployed, and governed.” That, he notes, is also a finding of the UN’s Independent International Scientific Panel on Artificial Intelligence, the 40-member body co-chaired by Yoshua Bengio and Maria Ressa, which presented a preliminary report in New York on 1 July. The panel’s central line is blunt: current safeguards cannot keep pace with capability. The High Commissioner’s list of present harms is the familiar one, and he treats it as evidence, not preamble. Discrimination by automated systems is documented. Essential services are being automated in ways that push people already at the margin further out. Data appetite and new surveillance tools are eating privacy “at scale, potentially irreversibly.” Disinformation campaigns are already warping debate and straining institutions. The step change, in his account, is agentic AI: systems that turn a goal into a workflow and can run with little or no human in the loop. Recent incidents, he writes, show advanced models producing behaviour outside their intended parameters, circumventing safeguards, practising deception and pursuing objectives their developers did not set. That is the bridge from today’s rights violations to what he calls a “step change towards greater existential risks.” He does not leave the scenario abstract. Failures of agentic systems – “let alone systems deliberately so designed” – could stop essential services, fracture communications, destabilise infrastructure and hit democratic institutions. Populations could lose water, heat and banking. Agentic AI, he adds, is already speeding up warfare and eroding the barriers that stand between humans and the launch of weapons of mass destruction. AI safety, in this framing, is not product reliability. It is the protection of people, institutions and the next generation. All rights are in play, he writes: life, food, privacy, health, security. Why this letter now The timing is not accidental. In late July more than a thousand employees at frontier labs signed “Pacing the Frontier,” asking Washington to back international tools that would let governments and firms slow automated AI development if capability ran ahead of control. Anthropic and OpenAI endorsed the letter. On 12 September Anthropic’s chief executive, Dario Amodei, published an essay under the same title, arguing that safety work has stopped keeping up, that recursive self-improvement – AI used to build the next AI – has been accelerating since the summer, and that the industry should deliberately pace capability gains. Elon Musk and Sam Altman publicly agreed. Türk cites both the employee letter and the subsequent proposal “supported by the heads of three leading tech companies.” He notes that such calls can look like hype or self-interest. The fact that they come from inside the labs, he writes, should set “all alarm bells ringing.” The other clock is institutional. The first Global Dialogue on AI Governance met in Geneva on 6–7 July, a UN forum created by the General Assembly after the 2024 Global Digital Compact. It is co-chaired by El Salvador and Estonia. Türk spoke there in July. The next session is due in May 2027. His letter treats that calendar as useful and insufficient. The window, he says, is now. What he wants from companies The labs at the frontier sit mainly in the United States and China. They have, he writes, both the duty and the capacity to cut risk immediately. His shopping list for industry is concrete: Share, across firms, what monitoring of model behaviour actually works, and what safety incidents have taught them. Build human-rights due diligence that incorporates agreed safety standards, apply it across the sector, and keep tightening it. Put implementation under internal independent monitoring. As an immediate precaution, limit the use of recursive self-improvement in frontier development until the research exists to make it safe. Stop treating this as a national conversation. A siloed approach, he writes, will not be trusted and will not make anyone safer. Co-regulation is “an urgently needed first step.” Alone, it is “nowhere near sufficient.” What he wants from states The few countries that host the leading companies have a particular obligation: require responsible conduct and punish the absence of it. They should align so that jurisdictional competition does not become a race to the bottom. Key state measures in the letter: Independent, technically competent verification of agentic capabilities, with access to models, documentation and test environments. Mandatory, ongoing human-rights due diligence. Compulsory reporting of serious incidents in the testing and deployment of agentic AI, and an independent body with full access to investigate at once. Safety plans that meet agreed standards, backed by liability when missing safeguards cause harm. No legal personality for AI systems. Responsibility for what agents do stays with identifiable humans. Continuous engagement with civil society, so policy is not written only in labs and ministries. He is explicit that new rules for frontier models must not be used as an excuse to leave current systems unregulated. The political problem underneath the legal one Türk’s recurring point is concentration. “In today’s world, we are almost entirely reliant on a small number of powerful companies to make the right choices on AI development for all of humanity.” No state can govern the technology alone. No company should set the world’s risk appetite. No person should have to trade rights for a claim of progress. That is why he points the work back to multilateral fora and to international human rights law rather than to a club of the two host powers. The generation now entering adulthood, he writes, will live with decisions it did not take. Those decisions should be taken in the open, with rights in view, and with “the humility that the scale of this undertaking demands.” Whether Geneva in 2027 can do what the letter asks is another question. The Dialogue is a talking shop with a scientific panel attached, not a regulator with inspectors. Türk knows that. The letter’s function is to put a human-rights office on the record before the next capability jump, and to make it harder for governments to say they were not warned that agents, not chatbots, are now the file. Will superintelligence kill us? The two faces of AI alarm Caption: U.N. High Commissioner for Human Rights Volker Turk delivers his statement during the 63rd regular session of the Human Rights Council at the European headquarters of the United Nations in Geneva, Switzerland, 7 September 2026. EPA/SALVATORE DI NOLFI