AI Highlights — the whole picture — 2026-10-31 (Sat) Daily News

The range of tasks entrusted to AI is expanding to include work, money, emotions, and drug discovery. Britain's NatWest was told it would cut 1,000 jobs and transform into an AI-driven bank, while US banks urged customers not to hand over their passwords to AI agents. The former indicates that the reorganization of work is progressing, and the latter indicates that the authority of the person entrusted to is not organized. Both moves are proceeding without a mechanism to see what happens after AI is introduced, or without a line being drawn as to who is responsible. In the following sections, we will look at employment, agency, nursing care and medical treatment, dependence on technology, and verification of authenticity. The focus of readers' judgments is shifting from how quickly to put things in, to how to check what they have put in, and how far they should keep it for themselves.

Behind the scenes of personnel reduction and growth, the nature of work will be reorganized

UK's NatWest reportedly plans to cut 1,000 jobs and transform into an AI-driven bank (JournalArta). In the United States, economic growth is reported to have accelerated, supported by consumption and AI investment (The New York Times). The former is a move by a single company to reduce its workforce and shift the focus of its operations to AI, while the latter is a move in which investment in AI leads to growth across the country. Although the scale is different, this shows that restructuring around AI is not just a matter of individual companies, but is moving in the same direction. For readers, the key factor for judgment is the fact that reductions and investments are progressing at the same time, rather than the number of people being cut or the size of the growth rate. When thinking about your company and your own work, you need to reconsider the assumption that introducing AI is a measure to improve the efficiency of a single department.

You can read other articles about how people perceive this reorganization. AP News took up the question of what to do if a company that made AI-related layoffs asks you to come back. (AP News) This question is premised on the fact that there may be a situation where the person who made the retrenchment wants the former employee to return, and that the person returning will need to assess the conditions and roles on their own. Experts explain that providing reliable AI requires a shift in thinking (CIO). I won't go into the details of this article here, but it shows that what is required of those responsible for implementation is expanding from making things work to ensuring trust.

There are also movements on the human resources side. California State University Bakersfield held a conference on the theme of developing AI talent (KBAK). Companies reshuffle their personnel, former employees are asked whether they can return, and universities provide training opportunities. These three movements reflect the fact that we are in a transition period where the nature of work is changing, and that who will play what role has not yet been determined. Readers will determine whether roles are being eliminated or newly created in their workplace, not by how quickly they are introduced, but by how well the role is defined and how it is developed.

If roles change, the next important question becomes who will check the results produced by the new person in charge and how.

The more you delegate, the more you need to draw the line between authority and judgment.

US banks urge customers not to hand over passwords to AI agents (Cleveland.com). When entrusting an agent with household finances and procedures, entrusting them with login information means giving them the same authority as the agent. What the reader should check is who is drawing the line between what the agent can and cannot do. Before considering convenience, the question is whether you can decide the scope of your authority yourself.

The same structure exists in the hiring field. AI-generated job ads increase the risk of citizenship-based discrimination (Bloomberg Law News). Even though the AI wrote the text of the ad, the responsibility for the results of posting the ad and selecting applicants remains with the company that posted it. Once you entrust the creation of a document, it does not mean that you are legally responsible for it. Additionally, the US Department of Justice is investigating Nvidia's acquisition of Hugging Face, according to a source familiar with the matter (Reuters). The question of who controls the foundation for AI development is now being examined by competition authorities. All three cases indicate who will assume authority and responsibility at the place where the task is entrusted.

Preparation on the part of the entrusting party becomes even more important when the agent himself changes. The recursive self-improvement survey (preprint on arXiv, not yet peer-reviewed) featured on its own site divided the process in which AI rewrites not only the model, but also prompts, memories, tools, and the next method of improvement into steps. We then organize the requirements for reliable improvements from three perspectives: ``Has things really improved?'' ``Are there any restrictions that need to be followed?'' and ``Is human supervision effective?'' It does not include new experiments and is an organizational framework that can be read in the abstract. This indicates that self-reported improvements should not be accepted at face value, but that it is necessary to separately confirm the maintenance of constraints and the effectiveness of supervision.

Even on the collaborative side, setups are becoming less fixed. The DynaFlow preprint (arXiv, not yet peer-reviewed, introduced on its own site) trains only the main agent, and writes instructions to the sub-agents and the flow of verification as a program for each task. The authors claim that it outperformed its peers on math, science, and search benchmarks. If the setup changes every time, a condition for operation is whether it is possible to record who verified what and in a way that can be traced after the fact.

The criteria for deciding whether or not to introduce the system will shift from how much responsibility is given to who will be responsible for confirming the scope of the entrustment and how. This line is directly connected to the question of how to verify actual results as AI enters specialized fields such as drug discovery.

Snuggling AI enters the field of medical treatment, nursing care, and loneliness

AI is beginning to enter a place close to the human mind and body. AI is becoming a part of elderly care (PYMNTS.com). An Indian psychiatric clinic has introduced emotion recognition AI to read patients' emotions (The Tribune). In the field of electrocardiograms, Cardiolyse announced a new workspace that uses AI to analyze and check 12-lead electrocardiograms (Yahoo Finance). What all three cases have in common is that the objects handled by AI have shifted from documents and numbers to records of people's emotions and hearts. For those considering implementation, it is not enough to just look at the accuracy numbers. Before implementation, it is necessary to decide who will make the final decision in the event of a misreading.

AI is also becoming a conversation partner at home and on the individual side. A survey found that Australia was the most searched country for "AI girlfriend" (IT Brief UK). It has been pointed out that while having an AI talking partner can be comforting, it can also deepen loneliness (Religion Unplugged). The same tool can be used as support for some people, and for others it can increase the distance between others. Because of this coexistence, it is not possible to judge whether a product is a good product or not based solely on the increase in the number of users or usage time. Additionally, there is a growing debate about whether AI can feel (VARIndia). The question of whether AI has emotions is also related to how users treat AI.

This subject is also covered in education and lectures. A Campbell University graduate gave a lecture on "Mental Health and AI" at a lecture in Paris in February (Campbell University News). This means that we have entered the stage where experts are publicly discussing the relationship between mental health and AI.

As the number of applications for these products increases, the methods to confirm whether they are effective or harmless, and the place of responsibility for the results will be questioned.

Technology dependence, sovereignty, and enclosure by industry

According to EUobserver, the EU and China are in advanced talks on AI, with the issue being whether they can cooperate beyond excluding Huawei and export restrictions. Regarding the UK, The Conversation covers the country's ``AI sovereignty.'' The question that will determine the realization of the concept is how to reduce dependence on foreign technology. Both articles show that there is a difference between being able to use AI and being able to support it yourself. For procurement personnel and managers, the decision to introduce a system is no longer based solely on performance and price. Before entering into a contract, it will be necessary to confirm which country's technology is providing the infrastructure and whether substitution will be effective if regulations or export controls change.

In drug discovery, funding and partnerships are moving first. AstraZeneca signs a deal worth up to $247 million with Abscix for AI-based antibody drug discovery (Dealroom). Because it says "maximum," it should be read as the upper limit if the conditions are met, rather than the full amount determined. Anthropic has acquired a biotech startup that uses AI to discover drugs (European Pharmaceutical Review). China's XtalPi is expanding its partnership in Singapore (Tech in Asia). Meanwhile, 10% shareholders of Tempus AI sold common shares worth US$17.59 million (Moomoo). Large contracts and acquisitions are occurring at the same time as major shareholder sales. It is not possible to tell the reason for the sale from these reports alone, but when looking at partnership announcements, it is better to look not only at the scale of the deal, but also who is involved and on what terms.

The same movement is also affecting agriculture. Eva announces AI-powered precision agriculture platform (Investing.com). Both drug discovery and agriculture involve operators with data and models moving into specific industries. The more difficult it is for providers to change, the more the problem of dependence mentioned earlier shifts from between countries to individual companies and on-site contracts.

The race to speed up implementation continues, but it has not yet been decided how the results will be verified and who will be held responsible when problems occur. The next question is how to confirm this and where responsibility should be placed.

Where to check authenticity

Dothan Civic Center is looking for local artists to create event posters, but no AI (WDHN). The criteria for this recruitment was not the quality of the painting, but who made it and how it was made. It is easy to write that AI is not allowed, but it is unclear from the materials used here how to confirm that the submitted work does not use AI. It is one thing to state the conditions and another to ensure that the conditions are met.

The other is research that looks for tampering that involves replacing only part of a long video (own website). If the video is tampered with by using AI to replace only a short scene of speech or sound while leaving the majority of the video genuine, it will not be detected even if the entire video is determined to be genuine or fake. A pre-peer-reviewed preprint published on arXiv created AV-FABLE, a long-form video benchmark with separate time axes for audio and video. Furthermore, they report that NFSR, a detection method that directly reads the internal features of a large-scale language model (AV-LLM) that handles audio and video, was able to extract evidence of tampering better than asking the language model (AV-LLM) verbally. However, what you can read here is only abstract and has not been peer-reviewed.

When you put the two side by side, you can see that even though the objects to be checked are different, the shape of the problem is the same. For posters, the question is whether the creator was a human or an AI, and for videos, it is a question as to which sections of the video are genuine. Neither can be judged just by looking at the completed product. If readers have been thinking about whether to use AI or not, they will now also look at who can confirm whether AI has been used or not, and by what means. What the video research shows is that it is not enough to judge the whole thing at once, and it takes time and effort to examine each time axis. The Civic Center's recruitment shows that a declaration of prohibition alone is not a means of confirmation.

The more scope is entrusted to AI, the more work remains after implementation to ensure that deliverables are authentic and that accountability is clear. In the next section, we will move on to the question of who is responsible for this confirmation and responsibility.

The topics in each section are different from employment, authority, nursing care and medical care, dependence on technology, and verification of authenticity, but what they all have in common is that the move to rely on AI is ahead of the means to verify results and safety. Headcount reductions and investment are progressing, agents are at risk of being given the same authority as the person in question, and they are now able to read emotions and electrocardiograms. It is one thing to set out conditions such as the country of dependence and whether the creator is human or AI, but it is another to confirm that they have been complied with. Just as companies are responsible for the results of texts created by AI, responsibility does not shift even if you leave it to them. Decisions from now on will depend not on the speed of implementation, but on how to verify the results, where to demarcate authority and responsibility, and whether or not you are prepared to support the infrastructure yourself.

Q1: Does your current workplace have a procedure in place to be the first to notice when the results of the assigned work are incorrect? Q2: Are the scope of authority given to AI and the final decision-maker in the event of an error determined in a document that everyone can confirm? Q3: If regulations or transaction conditions change and you can no longer use the AI platform you are currently using, are there any alternative ways to continue your business?

Finally, three questions

- Is there a procedure in place at the current workplace to be the first to notice when the results of the assigned work are incorrect? Q2: Are the scope of authority given to AI and the final decision-maker in the event of an error determined in a document that everyone can confirm? Q3: If regulations or transaction conditions change and you can no longer use the AI platform you are currently using, are there any alternative ways to continue your business? - Is the scope of authority given to the AI and the final decision-maker in the event of an error determined in a document that everyone can confirm? Q3: If regulations or transaction conditions change and you can no longer use the AI platform you are currently using, are there any alternative ways to continue your business? - If regulations or trade conditions change and you can no longer use your current AI infrastructure, are there alternative ways to continue your business?