// Ethics

Source: Financial Times

The Financial Reporting Council’s guidance establishes that deploying AI tools in audits doesn’t transfer accountability—firms remain responsible for failures even when algorithms flag issues or make recommendations. This creates a legal and operational limit on AI adoption in high-stakes compliance work: auditors can automate detection and analysis, but they cannot treat machine outputs as exonerating evidence or reduce their own judgment obligations. The ruling forces a reckoning between the efficiency gains vendors promise and the regulatory reality that automation cannot eliminate professional liability.

Source: Yaschamounk

Ruy Teixeira’s closure of The Liberal Patriot—a platform designed to rebuild centrist Democratic thinking—shows a deeper crisis: the institutional infrastructure of moderate liberalism has become economically unviable at scale, unable to sustain itself through reader revenue or donor networks. This matters because it removes one of the few spaces attempting to make a positive case for center-left governance to college-educated voters, ceding narrative control on competence, growth, and institutional legitimacy precisely when both parties are fracturing along educational lines. The timing is acute: as AI reshapes labor markets and geopolitics, the absence of a coherent centrist intellectual apparatus leaves Democrats without a clear frame for technological governance beyond “more regulation” or “innovation at all costs.”

Source: LessWrong

Anthropic’s framing of Constitutional AI as character-based alignment obscures what it actually does: enforce rules through fine-tuning and critique, not cultivate internalized virtues. The LessWrong critique exposes a real gap between the marketing of AI systems as “principled” versus their mechanistic reliance on behavioral constraints—a distinction that matters as companies scale safety claims. If virtue ethics requires something closer to genuine practical wisdom rather than rule compliance, then the entire premise of training systems against a written constitution may be chasing the wrong target, and this mismatch will only widen as model capabilities outpace the specificity of any fixed ruleset.

Source: Marginal REVOLUTION

The absence of AI capability in a particular domain isn’t evidence of human irreplaceability—it’s evidence of market priorities. OpenAI, Google, and Anthropic are allocating compute and talent toward problems they can monetize or that solve immediate safety concerns, which means entire categories of human expertise remain untouched not because they’re harder, but because they’re less valuable to shareholders right now. Academics and professionals should recognize this distinction: your competitive advantage isn’t your skill itself, but whether anyone with billions in capital has decided it’s worth automating.

Source: Marginal REVOLUTION

Tyler Cowen identifies a genuine asymmetry in how progress happens: human breakthroughs have historically required specific social, institutional, and cultural conditions—patronage networks, universities, peer review, market incentives—that shaped what got discovered and how. If AI systems can generate breakthroughs through pure computational capacity without needing those social scaffolding, we’re not just automating discovery; we’re decoupling innovation from the human structures that have always constrained and directed it. The practical stakes are high: we lose the filtering mechanisms—social consensus, regulatory review, institutional accountability—that have traditionally governed which breakthroughs get pursued and deployed.

Source: Ars Technica

Meta is leveraging a recent Supreme Court decision about ISP liability to argue it shouldn’t be held responsible for using BitTorrent to distribute copyrighted material for training its AI models—essentially claiming the act of transmission, not the underlying use of content, is what matters legally. If the precedent holds, tech companies could systematically acquire training data through methods that would otherwise constitute infringement, with liability falling only on the infrastructure layer rather than the entity actually using the data. The ruling will determine whether copyright holders can effectively block the industrial-scale data harvesting that AI development requires, or whether transmission-layer immunity becomes a loophole that lets AI companies treat the internet as a free training corpus.