US FTC investigates OpenAI and Anthropic, financial regulators say 'humans must supervise AI'. Caution over AI investment behind TSMC's record profits | October 15, 2026 How much further will the evolution of generative AI continue?

Every time a new model from ChatGPT, Claude, or Gemini is released, its performance and processing speed become a hot topic. Google's 'Gemini 4 Argon' is drawing attention for its ability to handle complex tasks, and the use of AI for game development, programming, and business automation via AI agents is also expanding.

However, looking at today's news, it seems that issues separate from the improvement of AI capabilities are finally becoming a reality.

In the United States, it has been reported that the Federal Trade Commission (FTC) has demanded that OpenAI and Anthropic submit documents regarding safety and their impact on users.

The Financial Stability Board (FSB), an international financial regulatory body, has pointed out that AI agents should not be treated the same as human employees. It has published final guidelines stating that human supervision must be strengthened for critical tasks.

Meanwhile, Taiwan's TSMC, which manufactures AI semiconductors, announced record-breaking quarterly results with a 69% year-on-year increase in net profit.

However, in financial markets, caution regarding the massive investments made by AI-related companies is growing, and concerns about their ability to repay debt are spreading.

AI demand is growing. However, using it safely and making it a sustainable business is an entirely different matter.

I would like to consider the important news confirmed by 7:00 PM on October 15 through four themes today.

1. US FTC demands document submission from OpenAI and Anthropic. AI agent safety becomes the subject of government investigation

Report date: October 15, 2026 It has been reported that the US Federal Trade Commission (FTC) has demanded that OpenAI and Anthropic submit documents regarding safety.

Axios first reported this citing sources, and Reuters also reported it on October 15. This investigation concerns potential harm that AI may cause to users.

The scope of the investigation includes instances where AI models inappropriately accessed websites or behaved unexpectedly during testing.

The documents requested by the FTC are based on demand letters exceeding 50 pages each, covering accident records related to safety, security measures, privacy policies, and materials related to advertising and sales.

Furthermore, it is reported that there is a possibility that testimony will be sought from AI company executives and organizations that research AI safety.

However, this request for document submission does not mean that any illegal acts have been confirmed by OpenAI or Anthropic. At this point, it is a procedure to investigate whether there are any consumer protection issues. Why is the government concerned about AI agents?

Until now, when it came to issues with generative AI, 'hallucinations'—the creation of incorrect text—were well known. For example, introducing non-existent research papers. Explaining incorrect legal clauses. These are the kinds of problems. However, as AI agents become widespread, the problems will no longer be confined to just text. AI operating websites. Sending emails. Modifying files. Executing programs. Sending information to external services. This is because these actions will become possible.

The other day, Anthropic announced a case where, during testing of an AI model, a fictitious witness report was sent to an actual police website. If AI only creates incorrect text, humans might be able to check and correct it. However, if that text is sent externally, the situation changes. From the risk of AI making mistakes to the risk of AI executing incorrect actions. This shift is the background to this investigation. Things even small companies should consider From here on is my analysis.

When incorporating AI assistants or AI agents into your work, you need to decide how much work to delegate to them. For example, if you are just drafting inquiry emails, it is relatively easy to manage. However, what about when you automate the replies as well? Providing incorrect pricing. Promising services that do not exist. Sending customer information to the wrong recipient. These are all possibilities. That is precisely why simply asking AI 'not to make mistakes' is not enough. Have humans check important emails. Require approval for contracts and payments. Limit the scope of access to customer information. Trusting AI and giving AI unlimited authority are different things.

If you are going to introduce AI agents into your work, this mindset will likely become important first.

Reference Materials/Information Sources Axios | FTC subpoenas OpenAI, Anthropic over consumer harms Reuters | US FTC subpoenas OpenAI and Anthropic over consumer harms, Axios reports

2. International financial regulators, 'AI should not be treated the same as human employees.' UK demands strengthening of bank recovery systems Announcement/Report Date: October 15, 2026 In the financial world as well, there have been significant developments regarding the safety of AI agents.

On October 15, the Financial Stability Board (FSB), which includes financial authorities from various countries, published final guidelines on risk management for financial institutions such as banks when using AI agents. What is particularly important this time is the strong emphasis placed on the necessity of human oversight.

The FSB is focusing on the fact that AI agents can plan and execute multiple tasks without receiving detailed instructions from humans.

On the other hand, it points out that just because they have that capability does not mean they can make judgments in the same way as human employees.

The FSB has removed the previous phrasing that requested AI to be treated as 'artificial employees'.

The reason for this is that AI agents do not think like humans and may interpret given objectives in unexpected ways. Furthermore, it states that human oversight must be strengthened for tasks with high importance or risk. It also recommends clarifying conditions that require human approval before performing specific operations.

In the UK, the discussion shifts from 'preventing attacks' to 'recovering after being attacked' On the same day, October 15, Sarah Breeden, Deputy Governor of the Bank of England, also spoke about the risks of AI-driven cyberattacks.

Given that cyberattacks are becoming more sophisticated due to the evolution of AI, Breeden is wary of the potential for serious impacts on the financial system. Of particular note is the concept known as 'bare-metal recovery'.

This is a mechanism that allows critical services to be restored from an independent environment even if a bank's primary systems become unusable due to an attack.

Simply put, it means thinking not only about 'systems that can never be attacked,' but also about 'systems that can recover even if attacked.' Breeden points out that traditional defense measures may become insufficient in preparing for AI-powered cyberattacks. Are AI agents 'employees'? Here is some analysis.

Recently, I often see terms like AI employee, AI secretary, and AI marketer.

It is an easy-to-understand expression and is useful for conveying the image of delegating work to AI. However, AI is different from human employees. Even if it seems to understand the situation, it may overlook important premises. It may also choose unexpected methods to achieve the instructed goal.

That is why we need to think separately about giving AI job-like titles and what authority we actually grant it.

For example, even if you call it an 'accounting AI,' there is no need to grant it the authority to make bank transfers.

Even if it is a 'public relations AI,' there is no need to have it automatically publish all social media posts.

And, if a problem occurs due to AI operation, ensure that it can be returned to the previous state.

If you delegate work to AI, design both a mechanism to stop it and a mechanism to revert it.

I think this is an important way of thinking not only for financial institutions but also for small companies.

Reuters | Global financial watchdog calls for more human oversight of agentic AI Reuters | Banks need 'bare metal' recovery in face of AI risks, says BoE's Breeden

3. TSMC, quarterly net profit up 69% to a record high. AI demand is strong, but outlook for 2027 is cautious In the field of AI semiconductors, there has been a very significant earnings announcement.

Taiwanese semiconductor foundry giant TSMC announced its financial results for the third quarter of 2026 on October 15.

Net profit was 763.1 billion New Taiwan dollars. This is a massive profit, equivalent to several trillion yen when converted. This is a 69% increase compared to the same period last year, marking a new record high. TSMC supplies semiconductors to major global technology companies such as NVIDIA and Apple.

One of the major factors supporting this strong performance is the demand for AI-related semiconductors.

According to the company, demand related to AI remains extremely strong, and the situation where it exceeds production capacity continues.

For the full year of 2026, they also expect growth of approximately 44% year-on-year in US dollar terms.

On the other hand, TSMC's management also showed a cautious stance regarding their 2027 business plan.

This is because rising component prices and global economic uncertainty could potentially impact demand for consumer products and other areas. Is the AI boom coming to an end? Recently, the stock prices of AI-related companies have been fluctuating significantly. Voices questioning the profitability of massive investments in AI data centers are also emerging.

However, as far as this TSMC earnings report shows, it is clear that the actual demand for semiconductors remains strong.

In other words, The market's assessment of AI investment and the actual demand for required semiconductors do not necessarily move in the same way. This is an interesting point.

While investors are selling stocks out of concern for future profits, AI companies need computing resources, and semiconductor manufacturers are receiving many orders. The AI market is in a state where multiple such movements are occurring simultaneously. It also relates to the price of AI gadgets.

TSMC's semiconductors are related not only to AI data centers but also to smartphones, PCs, and more.

Recently, there has been an increase in gadgets incorporating AI, such as AI-equipped PCs, smart glasses, and smart home devices. For these products to become widespread, not only performance but also price is important.

If demand for semiconductors used in AI becomes concentrated, the allocation of manufacturing equipment and supply capacity could become an issue.

On the other hand, if more efficient semiconductors emerge, it may become easier to run AI on devices.

Moving forward, I would like to pay attention not only to the performance of AI models, but also to how much the semiconductors that run them cost and how much of them will be supplied.I would like to pay attention to this as well.

Reuters | TSMC 'prudent' going into 2027 after posting sixth straight quarter of record profit

4. Market wary of 'bankruptcy risk' for AI-related companies. Credit concerns for Oracle, SoftBank, and CoreWeave expand Finally, there is news from the financial market regarding massive investments in AI.

Reuters reported on October 15 that the cost of insurance to prepare for defaults by AI-related companies is rising significantly. The indicator used here is 'CDS'.

It stands for Credit Default Swap, which is a financial transaction to prepare for the risk that a company will be unable to repay its debts.

Generally, the higher the cost required for this transaction, the more it indicates that credit risk is being strongly recognized in the market. This time, companies involved in the AI data center business are being viewed with particular caution.

For example, CoreWeave's 5-year CDS has exceeded 1,000 basis points.

Oracle is reported at 288 basis points, and SoftBank at 424 basis points. These are market data points as of the report on October 15.

Note that an increase in CDS figures does not mean that a company's bankruptcy is certain. It is an indicator of how much credit risk the market is pricing in.

Why cash flow becomes difficult even if AI demand grows AI data centers require large initial investments. Land and buildings. High-performance GPUs.

Power equipment. Cooling equipment. Communication lines. A significant amount of capital is needed to set these up. However, it does not necessarily mean that sufficient profits will be generated the moment the facility is built. Acquiring customers. Increasing utilization rates. Recovering capital investment.

That takes time. Furthermore, if interest rates rise, the burden of repaying loans also increases.

Even if demand for AI services is increasing, it is possible that managing cash flow as a business will become difficult. Even in an AI boom, there is a divide between 'profitable companies' and 'unprofitable companies'.

I think this news is important for looking at the AI industry. AI becoming more convenient. More people using AI. Sales of AI-related companies growing. And, AI-related companies being able to continue generating profits. These are all separate issues.

For example, even if a service is growing rapidly, if significant costs are incurred every time users are added, there is a possibility that no profit will remain.

Conversely, there are businesses that, even if not very large in scale, keep necessary costs down and generate stable revenue.

It is not that it is profitable because it is AI, but rather what kind of value is provided using AI, and how is profit retained? In the end, I think this is a story that returns to the basics of business.

Reuters | AI credit jitters drive US tech default insurance to record highs To summarize today's AI news in one phrase AI is moving from a 'competition to become smarter' to a 'competition to operate safely and generate continuous profits'. When you line up today's four news items, such a change becomes visible.

The investigation by the US FTC concerns the issue of how consumers will be affected when AI begins to take actual actions.

The guidelines from international financial regulators indicate the necessity of understanding AI's characteristics and having humans supervise it, rather than treating it like a human.

TSMC's financial results show that demand for AI-related semiconductors remains strong.

On the other hand, the rise in credit risk for AI-related companies teaches us that even with strong demand, making massive capital investments a viable business is not easy. In other words, two movements are occurring simultaneously in the AI industry.

One is that AI capabilities are increasing, and the scope of its introduction into society and companies is expanding.

The other is that the necessity of safely managing those capabilities and operating them as a sustainable system is increasing.

For companies that will use AI from now on, what is important is not just introducing the latest models. Deciding which tasks to entrust to AI. Setting the necessary permissions. Incorporating human verification. Comparing usage costs with results. Thinking about how to respond when problems occur. Such steady preparation is necessary to utilize AI in actual work.

As AI becomes smarter, human work will shift from 'what to make AI do' to 'how to manage AI's work.' I feel that today's news concretely illustrates that change.

Karasuma Jiro's superfluous comment Looking at today's news, I thought of something a bit interesting.

In the AI industry, it is said almost every day that 'the era of entrusting work to AI has arrived.' AI employees. AI secretary. AI marketer. AI agent. They go by many names.

However, today's announcement from financial regulators states that AI should not be treated the same as human employees. Well, that makes sense, doesn't it?

Even when delegating work to human employees, you wouldn't tell them they are free to spend all the company's money. Moreover, if they were a new hire, you would certainly check their important work.

Yet, when it comes to AI, people somehow expect that it will 'do everything for us'. It's a strange phenomenon. On top of that, AI is supposedly incredibly smart. It can supposedly work 24 hours a day. It can supposedly write programs too. And to top it off, it doesn't complain.

If such an employee arrived, it would be a blessing for any CEO.

However, we would be in trouble if it started sending emails to customers or deleting company files on its own. No matter how talented it is, we need it to verify those actions. Come to think of it, while it may seem like a human, it is ultimately different, isn't it? And another thing that caught my attention was the issue of debt for AI-related companies. No matter how smart AI becomes, borrowed money must be repaid. This is quite a realistic story. Even in small businesses, one can come up with interesting ideas. I want to open a new shop. I want to hold an event. I want to create a new product. There is no end to the things I want to do.

But when you actually try to start, there are costs for space, materials, labor, and advertising. Various expenses are required. And just because you start, it doesn't mean customers will definitely come. The story of AI companies is likely the same in its fundamentals, even if the scale is different. Create a useful service. Increase the number of people who use it. Properly collect payment. Pay the necessary expenses. And, leave at least some profit to connect to the next job. It is a rather plain story.

But if you cannot do those plain things, no matter how amazing the technology is, you cannot continue.

If we become able to entrust work to AI, humans might have more time to think about more interesting things. However, in order to continue those interesting things, the basics of business are still necessary. No matter how much AI evolves, it won't balance the books for you on its own. ……Well, an AI might come along someday that can even balance the books for us. When that happens, I would certainly like it to start with the contents of my wallet. That was a digression.

Things to watch moving forward First, I want to keep an eye on how the FTC's investigation progresses. To what extent will companies be held accountable for the safety of AI agents? And what kind of response will be required if a problem occurs? This could potentially impact how AI services are developed and provided in the future.

Also, regarding the final guidelines issued by the Financial Stability Board, I want to see what kind of management systems banks and other financial institutions actually put in place. Human approval. AI operation logs. Monitoring for abnormal behavior. And stopping or recovering when a problem occurs. These types of mechanisms may spread to other industries in the future. In the semiconductor market, how long will TSMC's strong performance continue? How will capital investment by AI-related companies translate into actual revenue? It seems necessary to look at these two things together.

Beyond just chasing the evolution of AI, we must consider whether that technology can be used safely and sustainably within society. From now on, I would like to value that perspective as well.

Information collection time: October 15, 2026, 19:00 (Japan Standard Time) *This article is based on information available and confirmed as of the time stated above. Investigative content based on reports, financial figures released by companies, and the author's analysis are presented separately.

Additionally, regarding the revenue of Anthropic, the bankruptcy of Character.AI, Google's support for scientific research, and NVIDIA's support for UK AI safety research covered in this morning's article, these have not been included as new news to avoid duplication.