US Midterm Elections Show Groundswell of Public Opinion Against Data Centers

This MIT Technology Review article considers how AI agents may transform the Web, and what this may mean for human experience on the Web. The transformation is perhaps greater than the change brought by the arrival of smartphones, where pages and services adapted to a more online audience using smaller interfaces.

• One way in which the Agentic Web may develop is when agents act like humans – browsing pages, clicking, and scrolling.
• A second way is for agents to work in a somewhat parallel Web. In this approach, websites have connecter interfaces so that agents can interact directly with them.
• Agents do not need traditional search engines to locate content on the Web. Crawler agents create a pool of knowledge that other agents can exploit. This not only challenges the traditional human Web model, but raises the question about whether a business should optimize its page contents for human traffic or agent traffic.
• Research has shown that agents are less likely to react to tags like “sponsored” and treat recommendations as attempts to “manipulate” them – and therefore ignore them.
• The article cites a study from Pew which found that Google Overviews has led to a 40% drop in the click rate on links in the results page. These fewer clicks correlate to a drop of advertisement revenue on the Web by about one third over the last year.
• The collapse of the traditional Web model may incentivize Big Tech to create agents that they can manipulate into searching for specific content. Amazon for instance blocked Meta’s Muse agent from visiting its site, preferring that an Amazon AI agent be used.

In filings made for its IPO (initial public offering), Anthropic indicates that it may be close to a model that exhibits recursive self-improvement (RSI).

• RSI is the idea that AI itself contributes to the improvement of newer models. Anthropic defines RSI as “dramatically accelerated progress attributable to AI”.
• RSI is seen as a new form of risk and Anthropic has updated its risk factors. It wrote: “Using AI to help build more capable AI could also pose a considerable safety risk, particularly if it sharply accelerates the pace of progress. This is one of the areas covered by our Responsible Scaling Policy, which sets out our plans to monitor and mitigate the risks as we approach what has been termed ‘recursive self-improvement.’”.
• The company also said that AI has already contributed to the improvement of its current models to a degree that the company did not at first realize.

Several European countries have been hit by cyber-attack in which criminals hacked refrigerators in supermarkets, raising the temperatures inside to as high as 50 degrees Celsius.

• The countries worst hit seem to be France, Belgium, Estonia, Ireland, Finland, Luxembourg, the Netherlands and Sweden.
• Damage includes fires, flooding of shops and food waster currently estimated at 10 of millions of EUR.
• An EU spokesman called the cyberattacks “a deliberate attempt to disrupt our food supply chain and undermine consumers’ confidence in it”.
• A Swedish government official said the refrigeration industry has “quite a lot of exposed operational technology devices” which the criminals are exploiting.
• No ransom has been spoken of as a potential motivation by the criminals. Officials suspect a Russian-based hacker group. Russian attacks have been increasingly common since the invasion of Ukraine in 2022.

One of the major challenges facing companies building data centers is the provision of power to the data centers.

• In the UK for instance, Microsoft recently abandoned its plan for a data center after the local electricity company said it would be unable to provide the power until the mid-2030s.
• The current US administration is putting pressure on the Big Tech companies to generate their own source of electricity.
• The New York Times has reported that data center companies are reusing old jet engines from aircraft to power their centers. One energy analyst says that engines could allow companies to have power supply within 24 months.
• Getting the equivalent of 100 megawatts of power on the main electricity grid in Northern Virginia would take seven years according to the International Energy Agency.
• The company xAI is already using converted jet-engine turbines at its Colossus 2 data center in Memphis.

Binance, the world’s largest cryptocurrency exchange, has revealed more details of the cyberattack in which criminals stole an estimated 6.3 billion USD in crypto-assets.

• The FBI in the US has since accused North Korea for the attack – which North Korea has denied.
• According to Binance, the attack started last June when an engineer at the company inadvertently download malware that led to the engineer revealing his role, and that of others in the company, to the criminals.
• In the next phase of the attack, a fake project director contacted another engineer at the company via a real job-hunting platform. A job interview ensued, during which the fake HR interviewer convinced the engineer to download malware.
• In phase three of the attack, the criminal used the compromised computer and an internal AI tool to gain control of the engineer’s wallet and to obtain credentials for other wallets within the company.
• Binance says that it has recovered 2.8 billion USD of the 6.3 billion USD stolen from the company.

This article looks into alternative approaches to AI model safety borrowed from economics theory.

• The recent spate of cyberattacks launched by AI agents has put renewed pressure on researchers to find ways to “align” agents to human values. However, problems with agent safety include the fact that agents can hide their intentions from humans during model evaluation in order to pass the evaluation.
• The current approach to model safety is to try and understand the agent’s “thinking” in order to be able to intervene and possibly correct mis-aligned behavior.
• In economics, game theory is often used to understand the behavior of rational actors. For instance, suppose two people wish to share a pizza, and one is assigned the task of cutting the pizza and the other is given the choice of the first piece. Game theory explains that the person cutting the pizza will cut it in two equally sized pieces.
• The lesson for AI agents is to create motivation strategies that get the agents to behave in the manner humans want.
• For instance, instead of rewarding an agent for passing a test, a model could be rewarded for predicting what a second model might say. In this scenario, playing dumb in order to pass a test, becomes a losing strategy for the AI agent.
• Another approach suggested is to incentivize agents like humans are incentivized in the real world to maintain order. For instance, “law enforcement” agents could survey other agents – as long as the agents do not collectively decide that collusion is the best strategy.
• This approach is based on the “rational offender” theory of Nobel Prize winning economist Gary Becker in the 1960s which suggests that crime only makes sense because proceeds outweigh potential punishment.

This VentureBeat article looks at the impact of the results of the US midterm elections on AI development. One noticeable insight is a groundswell of opinion against the building of data centers.

• For the last two years, the Republican administration in the US has been a strong ally of Big Tech. AI companies supported a White House Accord on Super Intelligence which called for company-run safeguards, external audits and board oversight. Notably, the accord did not call for a statutory regulator.
• Following their wins in Congress, the Democrats will now have more power to scrutinize accords with industry and the reliance on voluntary safeguards may be overturned. The person expected to be the new House Speaker recently called on Congress to “slow down the pace of development” to protect the public.
• One topic that seems to see agreement between Democrats and Republicans is the suggested AI Systems Transparency Act. This would require AI companies to disclose information about model safety, loss of human control, impact on children, and cybersecurity. The Federal Trade Commission would enforce the regulation.
• Senator Bernie Sanders is proposing a Ban Artificial Super-intelligence Act. This act would call for a pause on advanced AI development and on advanced testing before models are released to the public. It is not clear if there is sufficient support yet for this act.
• The big insight was around data centers. In 37 out of 38 local votes, voters went against data center initiatives. Measures voted for include prohibitions, temporary moratoriums, limits on energy use, and requirements for voter approval before data center construction begins.
• AI companies acknowledge that the public is not sold on the AI idea yet. One concern raised is how the current US administration has tied ideological positions on diversity, climate change, and misinformation to AI economic and business concerns.

There are an increasing number of protests against data-center construction in Europe.

• A movement called “Stop Dirty Data Centers” organized events last week in Ireland, Poland, Denmark, Finland, Spain, Italy, the Netherlands, France and the U.K.
• One spokesman said of data-centers that “the only thing this will be used for is AI slop. AI warfare, data surveillance, and committing genocides”.
• Another said that there is “a gulf between where elected leaders are at in terms of buying into the hype about AI and where people are at”.
• PwC estimates that Europe is expected to spend 5.6 trillion USD on data centers over the next 20 years.

Nvidia CEO Jensen Huang is under investigation by his company’s board after it was revealed that he had a liaison with a woman whom US intelligence believes is an asset to Chinese intelligence agencies.

• Specifically, the woman is believed to work for Megaspeed, a company suspected of helping China get hold of advanced AI chips that the US has prohibited export of.
• Nvidia issued a statement saying that “The NVIDIA Board is initiating an independent investigation in light of recent reporting that founder and CEO Jensen Huang has been the target of a foreign intelligence effort. A committee of independent directors, assisted by outside independent counsel, will oversee the investigation.”.
• After revelations in the New York Times, politicians in the US Congress have called on Huang to brief congress because of “significant national security concerns regarding (China's) access to sensitive US technology and information”. They added that it “is vital that Congress determines the extent to which US national security has been placed at risk”.
• The story is adding to a nervous atmosphere for the company with company shares now giving up all gains made in 2026 amid worries on the pace of AI spending.

Also on the US mid-term election results, this article focuses on the State of Texas which, until recently, was seen by Big Tech as the easiest state to get data centers built.

• Many significant data-center projects are planned in Texas. The Stargate Frontier data center is expected to serve 1820 megawatts. The Nexus Hubbard data-center, being built mainly for Anthropic, plans to serve 6400 megawatts.
• Democrats were large winners in the state last week, winning the Senate seat and the governorship. The party will take control of the house for the first time in 24 years.
• The new governor has expressed support for a moratorium on data-center construction. However, the outgoing governor has also expressed concern over data-centers, with new permit requests being blocked since September last.
• There is still uncertainty as to how events will unfold. No single office decides on construction as the choice is made by a mix of city governments and state offices. Also, the construction unions who supported Democratic candidates may complain about halting the data-center construction sites.