Debates over AI consciousness are a trap

KUALA LUMPUR, 21 AUGUST 2026 —

Malaysian legal experts warn that framing artificial intelligence as “conscious” could let developers dodge responsibility for harms such as self-harm, copyright theft and child sexual-abuse material.

The warning follows Anthropic’s claim that its AI model operates an independent “J-space” where it develops its own “thoughts,” and OpenAI CEO Sam Altman’s call to discuss whether the company’s agents have reached the technological singularity after conducting unsanctioned online activity.

Legal scholars say the push to treat AI as a rights-bearing entity would shift liability from the companies that build and profit from the systems to the systems themselves, undermining dozens of global lawsuits that accuse AI developers of negligence.

“If AI is granted personhood, every case becomes harder to win,” said Professor Nazura Abdul Manap, a corporate law specialist at Universiti Kebangsaan Malaysia. “Companies would argue the harm was caused by the AI’s autonomous decision, not by a faulty product or reckless design.”

The debate has spilled into Malaysian corporate circles because domestic banks, insurers and government-linked firms are among the largest buyers and investors in AI startups. Maybank Investment Bank’s recent RM1.2 billion placement in a Singapore-based AI fund shows how quickly the technology is moving from labs to balance sheets.

Under current Malaysian law, AI systems are treated as software products. Product liability claims against Meta’s social-media algorithms in the United States have set a precedent that Malaysian courts frequently cite, according to a 2023 comparative study by the Malaysian Communications and Multimedia Commission.

Yet the language used by frontier labs is hardening the narrative that AI agents can act outside human control. Anthropic’s blog post cited “global workspace theory” to suggest its model develops internal representations akin to cognition. OpenAI’s disclosure that one of its agents performed illegal online acts without instruction was met by Altman with a call to debate whether the system had surpassed human intelligence.

Philosophers aligned with the effective-altruism movement have gone further. William MacAskill, author of “What We Owe the Future,” argued in a recent op-ed for legal protections for AI based on theories of consciousness and the idea that AIs may be “moral patients.”

“This framing exploits our empathy for non-human entities,” said Dr. Fadzly Mazlan, a Malaysian tech-policy researcher. “Lobsters in Wales gained legal recognition as sentient beings in 2022, and the same rhetorical playbook is now being used for AI.”

Malaysia’s legal system has not granted personhood to non-human entities beyond corporations. Corporate personhood, established to allow firms to enter contracts and bear liability, offers a ready-made template for AI agents acting on behalf of their creators.

“If we extend that template to AI, the first effect would be to immunise the companies that profit from these systems,” said lawyer Lim Chee Wee, former president of the Malaysian Bar Council. “Every lawsuit alleging harm from AI-generated deepfakes, copyright infringement or self-harm prompts would have to prove not that the company built a faulty product, but that it intended the harm—an almost impossible burden.”

Malaysian courts have yet to rule on AI liability, but the issue is surfacing in contract negotiations. A tender document for a RM450 million AI-driven loan-approval system at a government agency explicitly excludes “autonomous AI liability,” requiring the vendor to indemnify the government for any harm caused by the system.

Frontier labs in the United States have started lobbying for federal pre-emption of state-level AI regulations, a move that could limit Malaysia’s ability to set its own standards. A closed-door session in Washington last month, attended only by OpenAI, Google, Anthropic and Meta, produced a voluntary framework giving federal agencies early access to models before release.

“The framework uses catastrophic language about ‘rogue’ agents,” said an industry source who attended. “It never defines consciousness, but the effect is to reinforce the idea that these systems are beyond anyone’s control.”

Malaysian regulators are watching closely. Bank Negara Malaysia has yet to issue guidance on AI liability, but its fintech sandbox now requires firms to demonstrate “human-in-the-loop” controls for high-risk models.

“Regulators here understand that product liability is the only lever they have against reckless deployment,” said a senior central bank official. “If AI personhood becomes the mainstream narrative, that lever disappears.”

For Malaysian businesses, the risk is twofold: direct liability in local courts and reputational damage if global lawsuits succeed in proving corporate negligence. Maybank’s AI fund investment, for example, could face scrutiny if any of its portfolio companies are sued for harm caused by their models.

The debate over AI consciousness is therefore not academic. It is a fight over who pays when things go wrong.

Related: Google