More than a century and a half ago, Charles Dickens wrote the iconic line, “It was the best of times, it was the worst of times”. Applied to AI governance today, it presents the perfect paradox: a technological golden age unfolding alongside unprecedented systemic risks.
The last two weeks in AI have been something else. In just a few days, the AI industry produced enough news to fill a month. A call from the head of a frontier AI lab to slow down the pace of AI development, endorsed within hours by some of his fiercest rivals with a lukewarm response from US President Trump, a mathematical result people will argue about for years, a bitter dispute over credit, a revolt by some of the world’s most decorated mathematicians and a researcher’s resignation that went viral.
Together, these look less like coincidences and more like an ecosystem under pressure. What does this mean for Pakistan? Pakistan will not train a frontier model. But we will live with all of them. Our students will use them. Our workers will compete with them. Our institutions will depend on them. The rules, however, are being written in Washington, Beijing, Brussels and inside the companies themselves. And the story is somewhat moving in the opposite direction for us.
In the last financial year, Pakistan’s IT exports hit a record $4.6 billion. But beneath that record, the entry tier of the industry is thinning. According to PASHA, entry-level hiring at major firms is down by about 25 per cent, employment among developers aged 22 to 25 has fallen by nearly 20 per cent since 2024, and entry-level freelance listings have dropped from roughly 15 per cent of the market to below 9.0 per cent.
Let’s explore what happened globally to understand what we can do locally. On September 8, OpenAI announced that an internal system, running some 10,000 agents over 88 hours, had cracked the Navier–Stokes problem, one of the seven Millennium Prize Problems of mathematics, which had remained unsolved for more than 90 years.
Then came the ugly part where prominent mathematicians Levent Alpoge and Tristan Buckmaster who had had been privately collaborating on a specific aspect of
the Navier–Stokes equations called OpenAI out. Buckmaster had used OpenAI’s Codex throughout for his calculations and asked whether its model had been trained on his sessions while attempting to solve the problem. He received no clear answer.
Whatever the truth of that particular dispute, it points to a much larger problem. AI companies increasingly own both the instruments researchers use and a stake in what those instruments discover.
The real backlash came on September 11, when 25 Fields medallists, the closest thing mathematics has to film stars, signed a declaration arguing that, while machines can produce correct results, turning famous mathematical problems into corporate benchmarks risks damaging the discipline. The concern was not simply that AI had solved something humans could not. It was the rush to announce results without the kind of write-up, attribution and accounting of prior work that mathematics has traditionally demanded.
For humans, solving a famous problem meant producing an insight that could be explained, debated and taught. A race to produce correct statements is not necessarily the same thing as producing knowledge. The process matters because it lets us ask the next question, often the one that leads to the next discovery or invention.
For years, when generative AI came for designers, writers, and musicians, we were told disruption was the price of progress and that AI’s real promise lay further up the ladder, but now even scientists and mathematicians are echoing the artists’ argument. Some say, however, that the argument will not hold up if AI makes ground-breaking research to eliminate disease.
That same week, Jacob Coxon, who spent three years working on research at OpenAI and then Anthropic, resigned with a warning that neither company was acting responsibly and that both were racing toward self-improving superintelligence. His post was reportedly seen more than 100 million times.
Days later, Dario Amodei, CEO of Anthropic, called on the industry to slow the pace of capability gains and committed Anthropic to giving outside evaluators employee-level access to its systems. Within hours, arch rival Sam Altman of OpenAI agreed and said OpenAI would do the same. Elon Musk also endorsed the idea.
The sudden harmony has inevitably bred suspicion. Do these companies know something about their own systems that is not yet in the public domain? Or is the real project to raise the cost of entry before anyone else arrives as competition while Anthropic prepares for an IPO. Also, if American labs slow down while Chinese labs accelerate, safety quickly collides with national security.
Yet both can be true: a company can genuinely fear what its technology is becoming and still recognise that regulation could strengthen its position. Whichever explanation is right, the consequences will play out elsewhere. And for countries outside the small group building these systems, that is the deeper problem.
Pakistan has a National AI Policy, a billion-dollar commitment and a target of three million AI-related jobs by 2030. But does any of this reach the tier that is actually disappearing? Reskilling mid-career engineers does not help a 23-year-old who cannot get a first job. The answer cannot be to tell an entire generation to compete with machines at being machines. We need to teach people not only how to use AI, but how to question it, judge it and decide what matters. The machine may produce the answer, but who develops the next question? Who decides which questions are worth asking? And who decides which problems matter for science, society and the future?
That is where robust AI governance comes in. It is not an abstract argument about distant existential risks; rather, it is a contest over who sets the agenda about AI that everyone else will eventually have to live inside.
We do indeed live in the best of times and in the worst of times. The trick, perhaps, is learning to tread between the two, with excitement and caution in equal measure.
The writer is a strategic comms, inclusion/digital policy adviser and teaches internet governance and tech policy at LUMS.
