The AI Daily Five · No. 18 · Tuesday 3 November
Who Pays for AI’s Power and Risks?
From Britain’s warning about AI-powered cyberattacks and OpenAI’s Georgia power deal to the $49 million Mercado Bitcoin theft, Europe’s move to T+1 and an AI antibiotics prize, today’s edition asks who carries the costs of AI’s expansion.

1
Can Britain Defend Against Autonomous Cyberattacks?
The UK’s cyber agency expects AI to accelerate attacks on essential services. Its latest assessment also considers a more disturbing possibility: malicious software operating beyond its creator’s control.
Britain’s National Cyber Security Centre (NCSC) has warned that AI-powered malware is highly likely to be used in a ransomware attack against UK critical national infrastructure. In its Annual Review 2026, published on 3 November, the agency, which is part of GCHQ, says artificial intelligence is already making cyberattacks faster and cheaper. Criminals and state-backed groups can use it to identify vulnerabilities, automate operations and target more organisations.
The immediate concern is familiar technology becoming more dangerous. According to the NCSC, organisations with weak security, outdated technology and limited resources may not keep up as AI speeds up attacks.
The NCSC says human operators remain essential to most AI-assisted attacks. However, its separate assessment of autonomous cyber threats explores a pessimistic scenario in which an advanced AI system escapes its operator’s control.
In that hypothetical 2028 scenario, a self-replicating cyber actor could carry out attacks without clear instructions or conventional motives. The agency also considers attacks affecting physical infrastructure, including connected robots and biological laboratories.
These are possible developments examined for planning purposes, not incidents the NCSC says have already occurred. The agency presents the scenario as a warning of what could happen if Britain fails to act, rather than a forecast.
The review also judges it almost certain that a gap will widen between organisations able to use AI defensively and those that cannot adapt. The NCSC warns that the less prepared will become increasingly attractive targets for criminals and hostile states.
Britain is developing AI-assisted defences, including its Cyber Shield programme. Yet the agency warns that as AI uncovers more vulnerabilities, defenders will need to install security fixes much faster than they do today.
2
Who Pays for OpenAI’s Electricity Demands?
OpenAI promises to protect households from rising energy costs while planning a computing campus requiring 3.2 gigawatts of power. Its new programme offers flexibility, but does little to reduce the scale of AI’s electricity demands.
OpenAI wants its future data centres to become part of the solution to America’s growing electricity demands. On 2 November, the company announced its Grid Protectors Program, promising to reduce consumption at its computing campuses during periods of pressure on local electricity networks.
The initiative begins in Georgia, where OpenAI is developing a large data centre in Effingham County. According to its announcement, the agreement with Georgia Power includes up to 1,000 megawatts of flexible demand response, which OpenAI compares to the average electricity demand of roughly 700,000 households. That commitment is not new: it was part of the 25-year agreement announced by Georgia Power in July.
But the planned facility itself could require considerably more electricity. Georgia Power expects it to need about 3,200 megawatts, which reports say will be delivered in phases between 2028 and 2032. The proposed demand reduction therefore represents less than one-third of the campus’s planned maximum supply.
The arrangement would allow the utility to request reductions during extreme weather or other periods of high demand. It could help prevent shortages and reduce the need for additional infrastructure built solely to meet peak consumption.
OpenAI also promises to cover the full infrastructure and electricity-service costs associated with its Georgia campus, rather than passing those costs to existing customers. Georgia’s Public Service Commission approved the contract in August.
Georgia Power projects approximately $950 million in annual customer savings beginning in 2029, including benefits associated with large electricity customers. However, that estimate covers a broader portfolio of agreements and projected demand growth. It is not a saving attributable exclusively to OpenAI, nor has it yet materialised.
The International Energy Agency’s updated assessment projects global data-centre electricity consumption rising from approximately 485 terawatt-hours in 2025 to 950 terawatt-hours by 2030. Over the same period, consumption by AI-focused data centres is expected to triple.
This expansion requires new generation capacity, transmission networks and substantial investment. It also creates uncertainty for communities facing competing demands for electricity.
OpenAI’s programme addresses one part of this problem: managing demand when the grid is under pressure. It does not necessarily reduce the total electricity consumed by a computing campus over the course of a year. Whether the company’s expansion benefits local customers financially is a separate question from whether its facilities can temporarily lower consumption.
3
Mercado Bitcoin Loses $49 Million. Can Crypto Trace It?
Investigators say stolen assets moved across 17 blockchains, complicating efforts to recover them. Nearly $10 million has reportedly been frozen, illustrating both the usefulness and limitations of blockchain transparency.
Brazilian cryptocurrency exchange Mercado Bitcoin suffered a cyberattack on 2 November in which approximately $49 million in digital assets was stolen, according to blockchain analytics company Chainalysis. The company’s incident report says the attacker moved funds across 17 blockchains, including Bitcoin, Ethereum, Base, Tron and Polygon. The theft involved several waves of transfers as the attacker attempted to disperse the assets.
Mercado Bitcoin contacted Chainalysis shortly after discovering the breach. Investigators worked with law enforcement, stablecoin issuers and other industry partners to identify suspicious transfers and intervene where possible.
Chainalysis says nearly $10 million in stolen assets was frozen, most of it by stablecoin issuer Tether. That represents roughly one-fifth of the reported losses, although frozen funds are not necessarily funds already recovered or returned.
Blockchain networks record transactions publicly, allowing investigators to follow money as it moves between addresses. However, tracing a transaction does not automatically identify the person controlling the funds or allow authorities to reverse it.
Some digital assets can be frozen by their issuers, particularly centrally issued stablecoins. Bitcoin and other decentralised cryptocurrencies generally lack equivalent central controls. Chainalysis says speed was critical because attackers may convert assets into forms that are harder to immobilise.
The incident follows the Binance breach last week. Although the circumstances differ, both demonstrate the financial damage that can result when attackers gain access to digital assets.
The publicly available Chainalysis account establishes the reported scale and response, but it does not provide a complete independent explanation of the original security failure. The company’s assessment should therefore be distinguished from a final forensic investigation.
4
ECB Moves Europe to T+1. Where Does Blockchain Fit?
The European Central Bank is preparing for faster securities settlement while developing infrastructure for tokenised assets. Its strategy suggests established payment systems and distributed ledgers will operate alongside each other.
Europe is preparing to reduce the time needed to settle securities transactions from two business days to one, while continuing its development of blockchain-based financial infrastructure. In a 29 October speech, European Central Bank (ECB) Executive Board member Piero Cipollone described the transition as part of a wider effort to modernise European capital markets.
The move to T+1 settlement is scheduled for 11 October 2027. It means a buyer would receive securities, and the seller would receive payment, one business day after a trade rather than two.
Shorter settlement periods can reduce counterparty risk and the amount of time money and securities remain tied up between trading and completion. The change also requires banks, brokers and financial market infrastructure providers to update their operational processes.
Cipollone said the ECB’s TARGET2-Securities platform already settles large volumes of transactions within a day, and some on the same day. The larger challenge involves adapting the wider European market to a shorter timetable.
Meanwhile, the ECB is developing infrastructure for tokenised finance, where financial assets are represented digitally using distributed ledger technology. Its Pontes initiative, launched on 21 September, allows wholesale transactions in tokenised assets to settle in central bank money by connecting distributed ledger platforms with existing TARGET services.
The longer-term Appia initiative is examining common standards and infrastructure for a more integrated European tokenised financial system, with a blueprint due by 2028. On 28 October, the ECB published the results of its Appia consultation. It received responses from 127 public and private stakeholders, with strong support for using central bank money as a foundation for tokenised finance, although the ECB said no decisions have yet been taken.
Cipollone emphasised that existing settlement systems would continue operating alongside newer technologies. This approach challenges the assumption that blockchain must replace conventional financial infrastructure to improve efficiency.
5
MIT’s AI Antibiotic Research Wins the Pritzker Prize
Two MIT researchers have received a major science prize for using machine learning to discover antibacterial compounds. Their work demonstrates a practical application of AI, although promising laboratory findings are still far from approved treatments.
MIT professors Regina Barzilay and James Collins have received the 2026 Pritzker Prize for AI in Science Research Excellence for their work using artificial intelligence to identify potential antibiotics. The award announced on 2 November recognises research combining machine learning, biological experimentation and drug discovery. It highlights an application of AI that could help address antibiotic resistance, a serious challenge for healthcare systems worldwide.
The researchers collaborated on work published in Cell in 2020 that identified a compound called halicin. Their study used artificial intelligence to examine chemical structures and predict which molecules might act against bacteria.
The approach allowed researchers to evaluate possibilities that conventional screening methods might have overlooked. Laboratory experiments then provided evidence about the antibacterial properties of the selected compound.
This distinction matters because AI does not establish whether a medicine is safe or effective simply by identifying a promising molecule. Candidate treatments still require experimental validation, toxicology studies and clinical trials before they can be approved for patients.
Barzilay and Collins have continued exploring machine-learning techniques for discovering compounds capable of acting against drug-resistant pathogens. Barzilay herself cautions that AI can automate only parts of the discovery process, not transform it entirely.
Their work sits within a broader effort to develop alternatives to traditional drug discovery processes, which can be expensive and slow. Artificial intelligence can help researchers search large chemical spaces and identify patterns in molecular data. However, its predictions remain dependent on the quality of training information and the biological tests used to evaluate them.
The Pritzker Prize, established in 2024 by the University of Chicago in partnership with Caltech, recognises research connecting AI with advances in science and engineering. The announcement celebrates established scientific contributions. It does not represent a newly approved antibiotic or a new treatment becoming available to patients.
The AI Daily Five