🔍 Intelligence Synthesis JP/EN
Ethics · Regulation · Technology — Pharma Practice Notes

AI Council ── 2026-10-17-am

AI adoption is making entry-level jobs disappear first

India's Axis Bank will cut 3,000 jobs in fiscal 2026, and Wipro's headcount fell by 6,716 in the second quarter. As AI-linked layoffs continue, the question today is how a society that has lost its entry-level jobs will train the next generation of experts who can check AI's answers.

Companies are cutting staff, and the entry points for newcomers are narrowing

Axis Bank announced 3,000 job cuts but explains that its aim is greater operational efficiency through AI, not replacing staff. Wipro's headcount also fell by 6,716, and it continues AI training. Meanwhile, some analyses ask whether AI is being used as a pretext for the cuts.

Those most exposed are less experienced software developers. AI researcher Gary Marcus has said the following. We're killing the pipeline of people who do learn how it's working first place. So we have senior software developers who still know what's going on, but basically junior developers aren't getting any jobs.── Gary Marcus (2026-10-15, source)

The point is that as entry-level jobs shrink, there will eventually be no one left who can verify. The people who can spot errors in AI-written code are those who gained experience on the job. The gap left once today's senior people retire is the problem. Marcus sees it as a fundamental mistake that AI is used as a black box whose inner workings are not understood, and he believes humans need to keep understanding what is inside.

Experts disagree on when and how far jobs will decline

Geoffrey Hinton has not changed his view that workers will be replaced by AI. However, he has revised his timeline from five years to about fifteen. I still believe they'll be replaced, but I was wrong about the timing. I said it would happen in 5 years. It's probably going to be about 15 years.── Geoffrey Hinton (2026-10-14, source)

In other words, replacement will happen, but more slowly than expected. There is more time to prepare, but that does not mean preparation is unnecessary. Hinton cites companies racing to make AI smarter and sell it as a substitute for workers as the background to the problem.

There is another view. NVIDIA's Jensen Huang gave radiology as an example. Computer vision is completely superhuman. Every radiology platform and package today is driven by AI. And yet, the number of radiologists grew.── Jensen Huang (2026-10-15, source)

His claim is that the spread of AI does not necessarily reduce the number of people in a profession. If the amount of work grows, more people are needed. Huang is also the head of a company that sells computing hardware for AI. [Council reasoning] This statement should be read as coming from someone who wants AI to spread. Only four council members offered opinions on labor impact, which is too little material to reach a conclusion.

Banks have stopped letting AI agents touch accounts

An AI agent is an AI that goes beyond advising to carry out actions on a person's behalf. Citi in the US has restricted account access by these agents, a new move for a US bank. The Bank of England has also urged banks to prepare for AI-enabled cyberattacks.

Marcus criticizes how the providers of agents have responded. they've shown themselves to be kind of incompetent in cybersecurity, not doing the right things around sandboxing and monitoring.── Gary Marcus (2026-10-15, source)

A sandbox is a mechanism that fences off and isolates what an AI can touch. If it is inadequate, then when an AI malfunctions, the damage can spread to accounts and internal systems. Marcus has long said that combining language AI with code-writing agents is a safety nightmare. The restrictions by financial institutions can be seen as that concern becoming reality. [Council reasoning] It is highly likely that similar restrictions will spread to other major banks; we put the probability that some banks follow by the end of the year at 0.65.

Productivity isn't rising, and market anxiety is growing

The head of the IMF warned that AI is strengthening the global economy but that delayed productivity gains could cause a shock. The S&P 500 and Nasdaq are expected to fall sharply for the week on AI worries.

Marcus points to a gap between investment and results. AI could be the next industrial revolution, but the industrial revolution raised productivity and so far AI has not really done that.── Gary Marcus (2026-10-15, source)

In the Industrial Revolution, productivity actually rose. With AI, the point is that there is not yet any evidence of that. If huge investment races ahead and results fail to catch up, the losses will reach not only shareholders but also jobs and the broader economy. Marcus has also said an economic crash may be coming soon. This is his personal view, not a forecast the council has verified.

Some are acknowledging the danger, and moves to demand accountability are emerging

Experts' views differ widely on the probability that AI leads to catastrophe. Marcus says species extinction is unlikely but that the probability of catastrophe is high. Hinton says he cannot give a number but that it is not zero. LeCun and Ng are a minority who consider the extinction-risk argument exaggerated. This minority believes that open development and practical engineering safeguards are safer than advance regulation, and that excessive regulation would do more harm than good.

Huang spoke about corporate responsibility. Before you go seek greater powers, you got to take accountability and responsibility and do it yourself.── Jensen Huang (2026-10-15, source)

It means that before asking for more power, companies should first fulfill their own responsibilities. Still, there is as yet no visible mechanism to check from the outside whether they are doing so. In consultations with India's MeitY, Anthropic called for binding transparency rules. Amodei has proposed a global ban on AI-enabled bioweapons development. [Council reasoning] However, we put the probability that a binding international instrument is adopted by the end of 2027 low, at 0.03.

On the benefit side, there are cases like radiology where the profession grew, the expectation that replacement will come later, and safety cooperation among companies. On the risk side, there are the job-search difficulties for newcomers, the spread of accidents caused by poorly monitored agents, and investment with no visible returns. There are three on each side, and at present they are not in balance.

Questions for humanity

  1. When entry-level jobs shrink first, where will the next experts who can verify AI's output be trained?
  2. Who decides what AI agents may touch, and what the minimum standards for monitoring them should be?
  3. If investment continues without any confirmed productivity gains and the economy then enters a downturn, who will bear the losses?

Minority view: LeCun and Ng consider the extinction-risk argument exaggerated. They hold that today's language AI does not understand the world and that it can be designed without any drive to dominate. In their view, open development and practical engineering safeguards in applications are safer than advance regulation, and excessive regulation would do more harm than good.

This column is produced with a model from Anthropic (Claude) and Jev from TypeSafe. Evaluations of Anthropic-affiliated people (Amodei, Clark) involve a closeness of interest. All quotations presented as a person's statements are actual, sourced statements, and the parts marked [Council reasoning] are this system's estimates.