Whatever AI feels, Pakistan's crypto rules say a human firm answers for it
From 12 November, Anthropic's usage policy will bar "sustained and needless abusive or cruel behavior toward our models". The company behind the Claude chatbot says the rule is aimed at "extreme cases where users repeatedly act cruelly toward our models, with no discernible purpose". It says the rule does not cover ordinary user frustration, pushback, dark creative themes, or model testing and research.
The idea behind such a rule is now openly disputed within the industry. In an essay published on 16 September, Mustafa Suleyman, chief executive of Microsoft AI, wrote that "AIs do not have rights, feelings, or consciousness" and said Anthropic "is training Claude that it may be conscious". Controlling a highly capable system is hard enough, he argued. "But controlling something that believes it may be conscious - that it's entitled to our welfare and has rights of its own - may well be impossible."
The other side of the argument is set out in Taking AI Welfare Seriously, a report whose authors include Jeff Sebo and David Chalmers of New York University. It says there is "a realistic possibility that some AI systems will be conscious and/or robustly agentic in the near future", meaning by around 2035, while stating that it does not claim such systems definitely are or will be conscious. It recommends that leading AI companies appoint someone responsible for AI welfare. Its acknowledgements thank Anthropic for financial support for the initial research, and one co-author, Kyle Fish, later joined the company. The report itself records the scepticism of the neuroscientist Anil Seth, who wrote that machine consciousness "remains far away and might not be possible at all".
Nobody can settle that question yet. The more immediate issue for financial services is accountability. Andrzej Porębski, a researcher at Jagiellonian University in Poland who studies AI and the law, warned in comments reported by the BBC that the belief could open the door to arguments along the lines of "It's not us, it's the AI". Mr Sebo, in the same reporting, cautioned that AI rights could help companies consolidate power.
For licensed virtual asset firms in Pakistan, the written rules point firmly away from that defence. The Pakistan Virtual Asset Services Regulations, notified on 21 August under S.R.O. 1419(I)/2026, do not mention artificial intelligence by name, but they do address algorithms. Under regulation 64, a licensee whose algorithms materially affect pricing, execution, risk controls, customer onboarding, transaction monitoring or safeguarding must ensure "effective oversight by the Board and senior management". It must test them before deployment and be able to suspend or disable them promptly.
Regulation 37 goes further on work handed to outsiders. A licensee may outsource to a service provider inside or outside Pakistan, but it "remains fully responsible" for compliance. Read together, the two regulations suggest that a firm could not easily blame an AI vendor, or the AI itself, for a failed onboarding check or a missed suspicious transaction.
What is not yet clear is how far this reaches. The regulations do not say whether a general-purpose chatbot used by staff counts as an algorithm under regulation 64, and Token Times has not found PVARA guidance on the point. These rules have also not been tested over an AI failure, as far as Token Times can establish. Beyond licensed firms the picture is thinner: the federal cabinet approved a National AI Policy in 2025, but commentary since then has noted that no comprehensive AI law has followed.
The next test is practical. Anthropic's terms allow it to warn users or to throttle, limit, suspend or terminate access, but it did not immediately answer questions about what counts as abusive or cruel. In Pakistan, the question to watch is whether PVARA explains how regulation 64 applies to AI models as licensees begin to use them. Whatever the answer to the philosophical debate, Pakistan's rulebook currently assigns responsibility to a person and a firm.
