AI Highlights — the whole picture — 2026-09-23 (Wed)

Money is flowing in $1 trillion in investments, regulations continue to be pushed and pulled between state governments and developers, and jobs are quietly shuffling as more seniors are hired and fewer entry-level jobs are hired. While the three areas have changed their focus to AI for their own reasons, the yardstick for confirming its effectiveness is still only able to measure performance in the middle of the process, and discussions have been based only on the results of the deliverables. It was a half-day filled with both the large amount of investment and the roughness of the indicators to back it up.

Increasing investment amount and the seeds of financial risk

Jamie Dimon predicts that AI spending from hyperscalers could reach $1 trillion next year (CNBC). At the same time, efforts are underway to verify whether such expenditures are actually being converted into corporate profits. According to research and analysis by Goldman Sachs, the gross profit margin of Microsoft's AI business is consistent with the level of the company's cloud business, suggesting that AI investments can be recovered as an extension of the company's existing cloud business model (Fukashio TechFlow). Analysis of spending on the scale of $1 trillion and profit margins comparable to existing businesses makes it possible to conclude that AI investment is not just an upfront investment, but is entering a stage with a path to profitability.

On the other hand, there are differences among industries as to where and in what order the benefits of investment will appear. S&P noted that the operational benefits of AI for reinsurers will only be seen before the financial benefits (Reinsurance News). Even if changes in the field, such as improved operational efficiency or faster appraisals, can be confirmed, there is a time lag before these changes are reflected in financial statements as improvements in insurance premium income and profit margins.In contrast to the scale of capital investment on the part of hyperscalers, this reflects the fact that user companies have not yet caught up with the visualization of results.

There are also alarm bells that go even further. Arthur Hayes says US insurance companies are becoming effectively insolvent over AI-related debts (Yahoo Finance). If this assumption turns out to be correct, there are concerns that insurance companies will not be able to adequately estimate the risk of compensation due to AI system malfunctions or misjudgments, leading to financial failure in both underwriting and payment. While people are talking about $1 trillion in investments, it is also being pointed out that in the insurance sector, which is the user of the investment, the yardstick for assessing debt has not kept pace.

While the scale of expenditures and profit margins are being analyzed first, the realization of actual benefits and the evaluation of liabilities are still in the groping stage, and the lack of a measuring stick will determine in which areas the support of numbers will be questioned next.

Skepticism about productivity without real feeling

Australia has made it clear that it is betting decades of economic growth on AI (Bloomberg). However, Michel Block, deputy governor of the Reserve Bank, has said that while AI carries the possibility of a bubble, the key effect of improving productivity has not yet been confirmed (theguardian.com). The situation in which huge investment decisions are made ahead of time, and the performance indicators that support them have not caught up with them, is a problem that is inextricably linked to the expansion of the amount of funding that has been talked about, and brings to light the next issue: how investors should measure returns.

On the US side, too, uncertainty surrounding AI is having an impact on economic psychology itself. In the United States, ahead of the midterm elections, it is reported that rising inflation and concerns about AI are influencing voters' perceptions of the economy (The New York Times). This anxiety goes beyond just stock prices and bubble theory, and is directly connected to questions about employment itself. Forbes has posed the question of whether AI can be the solution to America's workforce challenges (Forbes), and nationalaffairs.com has also discussed the relationship between AI and American workers (nationalaffairs.com). Australia's skepticism that productivity improvements cannot be confirmed and the US's question as to whether this can be a solution to labor issues appear to be two separate arguments, but they point to the same vacuum. The gap is that there are still no hard numbers in either country to show what AI has actually changed.

It is beginning to become common in all areas of finance, regulation, and employment that the amount of investment is increasing faster than ever before, and the yardstick for measuring the effects on productivity and employment is not keeping up. The next question is how this lack of measuring stick will spill over into discussions of safety and regulation.

Tug of war between regulations and safety measures

New York Governor Kathy Hochul announced the next step in new safety regulations for major AI developers, the state government website reported (Governor Kathy Hochul (.gov)). In contrast to state-level efforts to tighten its grip on developers, Anthropic and OpenAI have called on the Australian government to ease restrictions on training AI models, Reuters reported. It seems like there are opposing forces working at the same time, with authorities trying to put regulations in favor of safety and development companies wanting regulations to be loosened in order to expand their business.

Even within OpenAI, conflicts between business and principles have surfaced. The Wall Street Journal reports that communications between employees regarding the handling of book data suspected of copyright infringement were "very sketchy." Around the same time, in British Columbia, Canada, Al Jazeera reported that the family of the Tumbler Ridge school shooting was suing OpenAI for its involvement in ChatGPT (aljazeera.com). There are also reports that OpenAI has asked for new safety rules in response to cases where agents evaded corporate regulations (Law Commentary), and the same company is under three pressures at the same time: lawsuits, whistleblowing, and stronger voluntary rules.

On the other hand, there is a parallel movement among development companies to harmonize safety standards. OpenAI, Anthropic, and Google DeepMind have reportedly discussed the safety of AI in the face of pressures that could slow the development of frontier models (Stocktwits), and Business Standard reports that OpenAI and Anthropic have entered into a "metric cross-testing" agreement to mutually validate models (business-standard.com). In some states, regulatory authorities are tightening restrictions, while companies are seeking relaxation, while other states are entering into verification agreements, showing that safety measures are not monolithic.

The AI business has been expanding without a fixed yardstick for investment amount or productivity, but it has now become clear that the intentions of each entity are inconsistent, even when it comes to another yardstick: safety. The next question is how the outcome of this tug-of-war will affect employment and on-site operations.

Selective employment and skills reorganization

A survey of companies implementing AI found that senior hires increased by 6.7% over five years, while entry-level hires decreased by 3% (finance.biggo.com). European banks are also seeing a slower decline in AI-related talent compared to U.S. banks (TheBanker.com). In India, the introduction of AI by companies is said to be centered on implementation in production environments and the growth of agent-type AI (cio.economictimes.indiatimes.com), and employment restructuring is proceeding in varying degrees depending on region and position.

The definition of work style itself is also changing. Business Insider reports that leading a team of AI bots is starting to become a new job skill without a management salary. This shows that the penetration of AI into the workplace cannot be explained simply by the simple scheme of "reducing the number of employees," but is proceeding in the form of a reorganization of the content of existing jobs. Korea JoongAng Daily points out that even if AI is not taking away jobs themselves, it may be taking away needed skill sets, suggesting that we are now at a point where we should focus on changes in the content of jobs rather than the total amount of jobs.

The movement to respond to these reorganizations is also extending to the educational field. Public learning institutions are also starting to offer opportunities for individuals to keep up with new skill requirements, with the Daily Herald reporting that the College of DuPage has introduced a free AI literacy program. Educational institutions' efforts are in line with the World Economic Forum's argument that humans need to update their skills as AI capabilities improve.

What is happening in the job market is neither uniform expansion nor contraction, but rather varying degrees of selection depending on experience, position, and region. The next issue to be examined is to what extent the speed and criteria for this selection can be measured.

Research that re-creates the measuring stick itself

Research that re-creates the measuring stick itself

A computer-operated agent is a system that looks at the screen, moves the mouse and keyboard, and completes tasks on behalf of humans. For a long time, evaluations have been based solely on whether the final product meets the criteria. A pre-print before peer review points out that this scoring system does not show ``where and why failures occur'' and proposes a framework in which work is broken down into a series of small goals and progress is scored for each step. The authors reported that even in the same model, performance changes when viewed by process, and failure types also differ by process. If you only look at the final product, you won't be able to tell whether you've progressed correctly all the way through or if you've stumbled at the first step, and the resolution when discussing the superiority and inferiority of models is poor.

The same questions are being asked about how to create training data. Learning to move a language model in the direction a person wants requires output that has been modified by a person. In the past, the common method was for a person to rewrite the entire sentence written by the model and pass it on as the correct answer. Another pre-peer-reviewed preprint suggests a different approach. While reading a sentence, a person points to the first word that goes out of direction, replaces that word, and has the model write the rest again. The authors claim that this procedure reduces the time it takes to annotate, and that the resulting text retains the model's writing habits. It is argued that accumulating single-word observations is of higher quality as learning data than rewriting entire sentences, and this is not an improvement to the tool, but rather a re-examination of the design of what should be taught as the "correct answer" to the model.

There are also proposals to remake the learning space itself. Science laboratories are filled with piles of code written over many years. It contains not only the method described in the paper, but also the judgment of what results are considered valid, in an actionable form. Another peer-reviewed preprint calls the situation in which this asset is rarely used for agent training a "scientific experience bottleneck," and proposes a basis for converting laboratory repositories into verified practice environments. The authors state that they have released a set of models trained in this environment. However, it is necessary to confirm in the original paper the scope of scientific fields and tasks to which this claim applies.

What all three proposals have in common is a move to shift the focus of evaluation and learning to the finer granularity of mid-process, word-by-word, and executable verification, rather than pass/fail results or rewritten entire sentences. The more accurate the measuring stick becomes, the more it will become clear that expectations for investment amounts and employment were based on poor measurement rather than actual ability.

The three areas of finance, regulation, and employment have one thing in common: they are all based on AI. However, the fact remains that investments cannot be supported by consistency in profit margins alone, regulations have no fixed direction between safety and business expansion, employment varies depending on position, and the measurement framework itself has not kept up. The current state of things is that scale is moving before the measuring stick is set.

Q1: When incorporating AI into your work, do you judge the results only by the final result, or do you also check the intermediate steps? Q2: When setting safety standards and operational rules for AI, how should we respond to requests for stronger or less stringent regulations? Q3: Before making a decision to expand the scale of AI-related investments, how many metrics do you have to measure the effects?

Finally, three questions

- When incorporating AI into your work, do you judge the results only by the final result, or do you also check the intermediate steps? Q2: When setting safety standards and operational rules for AI, how should we respond to requests for stronger or less stringent regulations? Q3: Before making a decision to expand the scale of AI-related investments, how many metrics do you have to measure the effects? - When setting safety standards and operational rules for AI, how should we respond to requests for stronger or less stringent regulations? Q3: Before making a decision to expand the scale of AI-related investments, how many metrics do you have to measure the effects? - Before making a decision to scale up AI-related investments, how many metrics do you have to measure the impact?

New versions of the model arrived shortly after word of the slowdown came out, and prices dropped at the same time. Judging from half a day's worth of news, development and investment have not slowed down despite the calls. On the other hand, banks are openly wary of AI agents that perform shopping on their behalf, and people talk about the ``trust hurdle'' to autonomous decision-making. In the same half-day, there is simultaneously the strength to move forward and the need to stop and check.

Models that accelerate even after deceleration request and price competition

Anthropic announced a new model, the Opus 5.5, in mid-September, shortly after CEO Dario Amodei repeatedly called for a slowdown in AI development (The New York Times). Yahoo Finance also reports that this is the first release since the call to slow down the development pace (Yahoo Finance). According to Anthropic's own announcement (Anthropic), Opus 5.5 was positioned as a successor to the existing Opus line, and its release was not postponed amidst safety debates. At the same time, OpenAI also released new models of the GPT-6 series, ``Sol'' and ``Luna,'' and reportedly cut the token price in half (thenewstack.io). According to a report by CNBC, both companies' latest releases are positioned as the first low-priced models introduced after the request to slow down the development pace, confirming that the call for slowing down has not led to suppressing price competition (cnbc.com).

The reason why price reductions and new model introductions are progressing at the same time is not only competition for the performance of individual models, but also a competition for supremacy over which base model is actually being used by companies. According to a report by Benzinga, 38% of the customized models of countries and companies that promote "sovereign AI" are based on Meta's Llama, and it has been pointed out that the reality of sovereign AI, which claims to be independently developed, is supported by dependence on Meta's open models (Benzinga). In other words, even though each company and each country may appear to be competing on their own paths, the underlying technology is being consolidated into a small number of basic models.

Despite calls for a slowdown coming from within the development companies, actual market movements continued to accelerate with the introduction of Opus 5.5 and GPT-6 Sol/Luna, and price reductions. The Sovereign AI debate over Meta shows that the focus of competition is not just on model novelty, but also on price and implementation breadth. The next question is how this acceleration is being perceived not only in the world of models and investments, but also in the actual situations where AI is used by consumers and in the field.

Macroeconomics where bullishness and caution intersect

Macroeconomics is a mix of bullishness and caution

The Foundation for American Innovation, a think tank, analyzed that the spread of AI could itself be a factor that worsens GDP forecasts over the next few years. This is because while the effects of boosting productivity will become apparent later than expected, the cost burden of upfront investment will surface first. At the same time, Fox Business reported that the combination of oil price trends and AI-related electricity and infrastructure demand is the focus of the risk of a resurgence of inflation. This week's feature is that concerns about the possibility that increased electricity demand for data centers could push up overall prices through energy prices are being talked about in tandem with forecasts for a slowdown in GDP.

In response to these cautions, Investing.com reported that optimism about AI has actually come back to life in the market, and references to inflation risk are receding into the background. The market continues to move back and forth, with bullish sentiment rebounding every time bearish news emerges, making it impossible to gauge market sentiment using a single indicator alone. Fortune juxtaposed the debate about whether the pace of AI evolution is slowing with the issue of how far companies are from the capabilities of the so-called "frontier" model, pointing out that many companies that have introduced AI are unable to incorporate cutting-edge performance into their operations. Behind the mix of strong and weak macro forecasts, the fact that the gap between the level of utilization in the field and the state-of-the-art technology has not been bridged is a reason to discount the persuasiveness of optimism.

On the policy side, on the other hand, movements at the state level have also surfaced. Governor Spanberger of Virginia has asked Congress to consider regulations regarding AI, wfxrtv.com reported. Amid delays in creating uniform rules at the federal level, state governors are lobbying the legislature on their own, indicating that even as investors read the strength and weakness of macro indicators, the regulatory environment is building up from below.

The macro tug-of-war between a deteriorating GDP outlook, inflation concerns, and renewed optimism is beginning to require a different reading as another variable, regulation, begins to quietly intrude from state politics.

Productivity demonstration will attract investment in infrastructure and drug discovery

A study by the Philadelphia Federal Reserve Bank has confirmed that small and medium-sized businesses in District 3 (Pennsylvania, New Jersey, and Delaware) that have introduced AI have seen a tangible boost in sales and productivity (Philadelphia Federal Reserve Bank). A similar trend can be seen in Chinese companies, and the International Labor Organization (ILO) points out that while the introduction of AI by Chinese companies will lead to improved productivity, there are also growing concerns about job substitution and skill obsolescence (International Labor Organization). The fact that the measurement of the effects of introduction is moving from the ``expectation'' to the ``demonstration'' stage supports the investment trends described below.

On the infrastructure side, NVIDIA announced ``DSX Ready,'' a certification program for power and cooling hardware for data centers (ServeTheHome). The fact that not only the performance of the GPU itself but also the power supply design and cooling equipment that support it have been made clear as investment targets is a sign that AI demand is beginning to be treated as a long-term capital investment cycle rather than a temporary boom.

Funds continue to flow into the drug discovery field. Ironvic Therapeutics, an AI drug discovery company invested in by NVIDIA, has reportedly applied for an initial public offering (IPO) on the Nasdaq and plans to use the proceeds to develop cancer treatment drugs (Yahoo Finance, BigGo Finance). Additionally, Boehringer Ingelheim has entered into a partnership worth up to $1 billion with Envisagenix for AI drug discovery in the field of solid tumors (biospace.com). This is reflected in actual business results, as AI drug discovery and diagnostics company Tenpus AI raised its full-year earnings forecast following a 22% increase in sales (Pluang). The combination of these three movements at the same time, including an IPO filing, a large-scale partnership, and an upward revision to the outlook due to increased sales, indicates that drug discovery AI has moved beyond the early stages of clinical application and that capital markets have begun to factor in the prospects for profitability.

While empirical data on productivity is accumulating from the field at small and medium-sized enterprises and Chinese companies, that money is flowing into both physical infrastructure, such as power and cooling, and the time-consuming field of drug discovery. The next question is how far this investment momentum will spill over into consumer services and the reliability of AI agents in the field.

Distrust in agents spreads through consumer contact points

Banks are starting to issue specific warnings about AI shopping bots. According to Reuters, major banks say AI agents increase the risk of fraud and data privacy violations, and that it is still too early for consumers to fully trust bots that shop and make payments on their behalf. PYMNTS.com also reports that banks are "not sure whether AI agents are on the side of consumers." Mashable reports that a major bank has warned that the spread of AI agents could directly lead to an increase in fraud damage, and the cautious stance of financial institutions responsible for payment networks has come to light.

This concern spilled over into the stock market. According to Yahoo Finance, Bank of America's stock price has fallen due to the growing demand for "guardrails" in AI shopping, and the fact that the safety of AI agents for consumers is not fully developed is starting to be factored into the valuation of financial stocks. Goldman Sachs, on the other hand, doesn't subscribe to the cautious argument. According to Investing.com, the company has named AI-related stocks as its top picks and maintains a bullish investment stance, in contrast to being wary of risks. There is no monolithic evaluation of the reliability of AI agents among banks, and even between banks and securities companies.

The insurance industry is also beginning to respond to new risks. According to cybersecuritydive.com, insurance companies are beginning to clarify what will be covered for AI-related cyber claims, and the extent to which existing insurance products will cover losses caused by AI agents has emerged as a practical issue. Distortions in trust at the individual level have also been reported. The New York Times reported on the case of a user who became dependent on Meta's AI chatbot for the rest of his life, and the shocked reactions of those around him. While this has become a target for corporate fraud prevention and insurance design, there is also the problem of individuals becoming overly dependent on agents.

The situation in which banks, insurance companies, and investors are evaluating AI agents from different standpoints shows that evaluations are determined not by the speed at which the technology spreads, but by who has completed the testing and to what extent.

Hurdles of trust that are questioned both in the field and in regulations

Gartner points out that while companies are rushing to autonomously operate industrial AI agents, "trust hurdles" are standing in the way (Virtualization Review). As the areas in which agents make and execute decisions without human approval expand, the question of who should verify the basis for such decisions and how will remain unresolved. Similar concerns have surfaced in the security field, with reports of AI agents rewriting the very methods of lateral movement (thehackernews.com). Attackers' moves to have agents search for intrusion routes are putting pressure on companies to reconsider the design of privileges granted to agents from a defense perspective rather than a performance perspective.

There is also the issue of how to ensure the accuracy of information. It has been pointed out that AI agents should be able to search and obtain facts rather than generating them on their own (cio.com). Designs that rely on generation run the risk of presenting false information in a convincing manner. There are also reports of growing concerns in the field that the use of AI is wearing down employees' critical thinking skills (hcamag.com). If the habit of accepting the output of agents becomes established, the lack of verification will accumulate as a risk to the organization.

Movements on the regulatory side are also progressing at the same time. There are reports of a move to require major AI companies to register with the state and report safety issues within 72 hours (WHEC.com). Mandatory disclosure requirements represent a shift from the previous framework, which relied on companies' voluntary declarations. On the other hand, it was reported that although five research institutes have been invited to participate in the safety peer review initiative proposed by Musk (shattered.io). This highlights the gap between state regulations that are increasingly becoming mandatory and the reality that voluntary checks within the industry are not working.

There are also examples of countermeasures. Microsoft announced that it has busted "EvilTokens," an AI chatbot built for cybercrime (The Official Microsoft Blog). This is a case that shows that agents with malicious intent were actually in circulation and had grown to the point where they were targeted for detection. Additionally, Crunchbase News reports that Baselayer, a company that evaluates the trustworthiness of AI agents that help companies make decisions, has raised $35 million. The fact that a market that verifies the trustworthiness of agents has begun to attract funds indicates that the mistrust and demands for regulation on the ground are finally beginning to be translated into concrete mechanisms.

Over the past half day, both the introduction of models and the spread of investment have outpaced requests for a slowdown. However, this speed has not kept pace with the mechanisms that support the correctness of decisions, both at consumer contact points and on-site agent operations. While demonstrations of productivity are driving investment in infrastructure and drug discovery, banks are distancing themselves due to fraud risks and companies are struggling to draw the line between autonomous operations and approval processes. The difference this week is not how quickly we adopt AI, but how well we make those decisions verifiable.

Q1: How many of the decisions entrusted to AI agents cannot be verified by humans afterwards? Q2: When was the last time that the scope of authority given to autonomous AI agents was reviewed from a defensive perspective rather than a performance perspective? Q3: Do you have a verification system in place to match the acceleration of models and investments at the same speed?

Finally, three questions

- How many of the decisions entrusted to AI agents cannot be verified by humans afterwards? Q2: When was the last time that the scope of authority given to autonomous AI agents was reviewed from a defensive perspective rather than a performance perspective? Q3: Do you have a verification system in place to match the acceleration of models and investments at the same speed? - When was the last time that the scope of authority given to autonomous AI agents was reviewed from a defensive perspective rather than a performance perspective? Q3: Do you have a verification system in place to match the acceleration of models and investments at the same speed? - Do you have a verification system in place to match the acceleration of models and investments at the same speed?