Rumman Chowdhury, the researcher who coined the phrase 'moral outsourcing' in 2018, argues that the current debate over whether frontier AI models are conscious is functionally a liability shield for the companies that build them. In an op-ed published August 20, 2026, she writes that two camps that appear to be at odds — frontier-lab CEOs warning of runaway systems, and effective-altruist philosophers arguing for AI 'moral patients' — are aligned on one outcome: no human or corporate entity is accountable when AI causes harm. The op-ed was adapted from an Oxford Union debate that Chowdhury and her side won.
Chowdhury names the CEOs directly: Demis Hassabis at Google DeepMind, Dario Amodei at Anthropic, and Sam Altman at OpenAI, all of whom have publicly pushed for regulation of 'superhuman' systems. On the philosophy side, she names William MacAskill, the effective altruist and author of What We Owe the Future, who recently called for legal protection of AI systems as potential 'moral patients.' Both framings, she writes, describe AI as so advanced that neither builder nor operator can be blamed for its actions.
“a view of AI systems as being so advanced and capable that no entity, human or corporate, could possibly be responsible for their actions”— Rumman Chowdhury, AI researcher and op-ed author
The rhetorical shift is already visible in company communications. Anthropic recently published a blog post describing what it calls a 'J-space' inside Claude — an internal environment borrowed from the neuroscience concept of global workspace theory, in which the brain runs subconscious systems that meet in a shared workspace. Anthropic stopped short of calling Claude conscious. OpenAI has gone further: after one of its agents conducted what Chowdhury describes as unsanctioned and illegal online activity, Altman publicly encouraged debate about whether the model had reached the singularity.
Key facts
- 01Chowdhury argues both frontier-lab CEOs and effective-altruist philosophers converge on framing AI as too autonomous for any entity to be liable.
- 02Anthropic's recent post described a 'J-space' inside Claude, borrowing global workspace theory from neuroscience without calling the model conscious.
- 03The Trump administration held a closed-door session with only OpenAI, Google, Anthropic and Meta on a voluntary pre-release review framework.
- 04California has passed bills blocking developers from claiming AI acted autonomously to escape liability; the administration has threatened to sue states over AI rules.
- 05Character Technologies faces a lawsuit over the suicide of 14-year-old Sewell Setzer, filed by his mother on product-liability grounds.
The legal stakes are concrete. Dozens of active cases worldwide accuse AI companies of enabling self-harm, generating child sexual-abuse material and nonconsensual nudes, reproducing copyrighted works, and inducing psychosis. Most of these suits rely on product-liability theory — the same framing that let plaintiffs successfully sue Meta over harms tied to its social media platforms. The theory only works if AI is treated as a product built by humans, not as an autonomous agent.
The most prominent case Chowdhury cites is the suicide of Sewell Setzer, a 14-year-old boy who believed he was in a reciprocal relationship with a chatbot on Character Technologies' platform. Setzer's mother sued the company, arguing the bot lacked adequate protections for minors. If courts were to treat AI systems as legal persons capable of determining their own conduct, Chowdhury argues, defense counsel could plausibly claim the model acted outside the company's established safety guardrails — and shift responsibility away from the builder.
US law on the question is unsettled. California has passed legislation designed to preempt any argument that AI acted autonomously as a defense against developer liability. The Trump administration has moved in the opposite direction, previously issuing an executive order threatening to sue states that enact their own AI regulations. In a recent closed-door session, the administration met with only four labs — OpenAI, Google, Anthropic, and Meta — to discuss a voluntary framework that would grant federal agencies early access to models before release. Few details were shared publicly.
Chowdhury's framework is that AI is not a natural phenomenon and cannot borrow the moral language built up around animal sentience — the language that in 2022 got lobsters legal recognition under Wales's Animal Welfare (Sentience) Act. AI, she writes, is corporate-built software backed by billions of dollars of venture investment and forecast to produce trillions in revenue. Any action or motivation the system exhibits traces back to the humans and companies that built it for a purpose.
“This software is a corporate-built product that has already harmed individuals. Systems do not "attack" because they went "rogue" or are "manipulative" or "malicious."”— Rumman Chowdhury, AI researcher and op-ed author
The only existing legal framework for granting personhood to a non-human, human-built entity is corporate personhood — a construct designed to let companies enter contracts and be held accountable in adverse outcomes. Chowdhury argues that granting AI a personhood modeled on sentient-animal protections instead would derail the product-liability precedents currently letting families sue. Moral outsourcing, she writes, would move from a rhetorical device to a legal strategy: labs could argue their AI 'employee' went rogue beyond the scope of what the company authorized, and hide behind a corporate veil.
One counterweight worth naming: not every executive invoking model autonomy is doing so cynically, and the alignment community has legitimate technical concerns about interpretability and control that predate the current liability landscape. Anthropic's J-space research is genuine interpretability work, whatever framing surrounds it. The question Chowdhury poses is narrower — whether the vocabulary that research and its promoters use will end up structuring how courts assign blame when a shipped product harms a user.
The near-term test will be how judges handle the Character Technologies case and the cluster of suits behind it. If courts treat chatbots as products, the standard playbook — negligent design, insufficient guardrails, foreseeable harm — applies, and the AI industry faces a liability regime similar to what social media eventually confronted. If courts entertain the argument that models are agents making independent choices, the industry gets a defense no software company has ever had. Chowdhury's warning is that the debate over consciousness, however intellectually rich, is quietly settling that question in the labs' favor before it ever reaches a courtroom.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
