Inside AquAI - Bring AI to every drop

⚙️ A new Water Environment Federation (WEF) report published 24 April confirms AI has shifted from pilot programs to frontline water utility operations, with human-in-the-loop oversight named as the non-negotiable condition for maintaining public trust.

🌊 Jacobs launched Flood IQ on 21 April, an AI platform that fuses machine learning with real-time sensor feeds across drainage, rainfall, and coastal systems to help cities and utilities move from static flood models to continuous operational intelligence.

🤖 OpenAI released GPT-5.5 on 23 April, its most capable model yet, built for autonomous multi-step work and described by OpenAI President Greg Brockman as "a big step towards more agentic and intuitive computing," with early enterprise testing showing a measurable step-change in hallucination resistance.

⚖️ The EU AI Act high-risk AI obligations take effect 2 August 2026. With less than four months remaining and a political trilogue scheduled for today, 28 April, the window for water utilities to act is closing fast.

💧 AI data centre projects under construction in Arizona could spike the state's annual water consumption by 67% to five billion gallons, according to Bloomberg Intelligence modelling published 27 April, as Colorado River supplies face deepening climate-driven shortfalls.

Source: Machine Learning Paradigms in Natural and Engineered Water Systems: From Proof-of-Concept to Trustworthy Deployment, Water Research, Elsevier, April 2026. Link: https://www.sciencedirect.com/science/article/abs/pii/S0043135426006147

The conversation in this sector has moved on. The question is no longer whether machine learning (ML) works in water. The question is whether it can be trusted at scale.

This comprehensive review, published in Water Research this month, maps how ML models are transitioning from experimental research into operationally deployed systems across natural water bodies and engineered infrastructure. Covering methods including deep neural networks, physics-informed models, and random forest algorithms, the authors examine real applications ranging from water quality prediction and anomaly detection to pollution source identification and treatment process optimisation.

Key findings:

Why it matters: the review does not simply celebrate AI's potential. It names the conditions under which deployment will fail. For utility managers evaluating vendor proposals and for engineers scoping pilots, this paper is a useful reality check on what makes the difference between a research outcome and a working operational system.

Source: Southern Water Expands AI-Driven Pump Monitoring with Samotics, Smart Water Magazine / World Pumps, December 2025. Link: https://smartwatermagazine.com/news/samotics/southern-water-uses-samotics-ai-prevent-equipment-failures-and-save-costs

What happened: Southern Water, which manages 39,973 kilometres of sewer pipes and 3,476 wastewater pumping stations across southern England, entered a six-year £7 million framework agreement with Samotics in September 2025 to roll out the SAM4 condition monitoring system across critical assets. The technology uses electrical signature analysis (ESA) to continuously monitor pumps by capturing up to 20,000 data points per second, with AI interpreting deviations from normal current and voltage patterns to classify specific fault types, including misalignment, bearing wear, airlocks, cavitation, and blockages. A one-year pilot across 637 pumps detected 63 equipment failures before they caused service disruptions, helping avoid an estimated £5 million in damage and regulatory penalties. Currently, 1,458 assets across approximately 600 sites are monitored, with expansion to 3,500 devices planned through 2026.

Why it matters: the results address the two objections most utilities raise when evaluating predictive AI for wastewater infrastructure: accuracy and false positives. The SAM4 system delivers 97% alert accuracy with fewer than 1% false positives, achieved through human-in-the-loop validation. Southern Water also reported a 40% reduction in internal flooding and a 15% reduction in external flooding linked to blocked sewers following the rollout. For utilities managing dispersed, hard-to-access pumping assets under tightening environmental permit conditions, this case demonstrates that the technology is mature, the business case is quantifiable, and the deployment model is scalable.

Sources: EU AI Act implementation timeline, artificialintelligenceact.eu | EPA WRAP 2.0 press release, 16 April 2026, epa.gov | DTA AI Policy v2.0 update, dta.gov.au

Three developments water professionals need to act on before the end of June.

EU AI Act: the high-risk AI obligations under Annex III take effect 2 August 2026. Water supply systems are explicitly listed as critical infrastructure under the Act, meaning AI deployed in process control, demand forecasting, or network monitoring is almost certainly in scope. Obligations include documented risk management systems, technical documentation, conformity assessments, human oversight protocols, and EU database registration. Penalties reach up to €15 million or 3% of global annual turnover. The European Commission's Digital Omnibus proposal could push some deadlines to December 2027, but compliance experts are unanimous: do not rely on that extension. August 2026 is the date to build toward.

EPA WRAP 2.0: released 16 April 2026, the US Environmental Protection Agency's (EPA) Water Reuse Action Plan 2.0 names data centre cooling as a strategic new driver of water reuse investment for the first time in federal policy. WRAP 2.0 is not a regulatory mandate, but it directly repositions wastewater utilities as economic development infrastructure. Utilities that can supply treated effluent to data centre cooling systems gain a new revenue stream, relieve pressure on drinking water supplies, and make their region more attractive for technology investment. EPA has committed to work with states on streamlined permitting for this pathway, meaning utilities that engage now will help write the rules.

Australia, Digital Transformation Agency (DTA): the first new mandatory requirements under DTA AI Policy v2.0 take effect 15 June 2026, seven weeks from now. For government business enterprise water utilities, this is a compliance deadline, not a recommendation. Requirements include designated accountable officials, completed AI impact assessments for any production AI system, risk registers, and publicly accessible AI transparency statements. Utilities that have moved tools into production without this governance infrastructure are already behind.

Source: Introducing GPT-5.5, OpenAI Blog, 23 April 2026 Link: https://openai.com/index/introducing-gpt-5-5/

OpenAI released GPT-5.5 on 23 April, describing it as the company's most capable and intuitive model yet. Unlike previous releases focused on response quality, GPT-5.5 is designed around autonomous task completion: it takes a complex, multi-part task, plans a path through it, uses tools, checks its own work, and keeps going until the job is done. Key capabilities include agentic coding, computer use, data analysis, document creation, web research, and native support for Model Context Protocol (MCP) integrations. The model matches GPT-5.4 on per-token latency while using significantly fewer tokens on complex workflows, making it more efficient at scale. Early testing at regulated institutions, including the Bank of New York, highlighted a step-change improvement in hallucination resistance, which is the reliability characteristic that matters most for high-stakes operational environments. API pricing is $5 per million input tokens and $30 per million output tokens, with a one-million-token context window.

For water utilities and engineers, the practical entry point is document-heavy compliance and reporting work. GPT-5.5 can absorb large technical documents, extract relevant obligations, compare them against existing operational data, and produce structured compliance summaries. For utilities already running AI-assisted workflows, its MCP compatibility means it can connect to internal SCADA records, asset management systems, and laboratory data via the same protocol that multiple enterprise vendors are now building toward. The improvement in hallucination resistance is particularly significant for regulated applications where an incorrect AI output is not a minor inconvenience but a potential compliance or safety issue.

Tool: SAM4 by Samotics Link: https://www.samotics.com

Overview: SAM4 is a hardware-software platform that monitors industrial pumps and rotating equipment by continuously analysing current and voltage signals at up to 20,000 data points per second. AI interprets deviations from normal electrical patterns to classify specific fault types, including blockages, airlocks, cavitation, bearing wear, and misalignment, before they escalate into failures. The system operates remotely via a cloud dashboard and is specifically designed for submerged, inaccessible, and difficult-to-inspect assets, which describes most of the pump stock in wastewater networks.

Key features:

Category scores (1 to 5):

Overall comment: SAM4 is a mature, field-proven technology with a quantifiable business case. Its strength is in hard-to-reach or submerged assets where conventional vibration sensors cannot be fitted. It requires hardware installation and vendor engagement, so it is not a quick trial, but for any utility managing a large submersible pump estate under tightening environmental compliance pressure, the Southern Water results make this worth evaluating seriously.

Source: Arizona's AI Boom Is Under Threat by Dwindling Colorado River, Bloomberg, 27 April 2026 Link: https://www.bloomberg.com/news/articles/2026-04-27/arizona-s-ai-boom-is-under-threat-by-dwindling-colorado-river

AI data centre projects currently under construction in Arizona could push the state's water consumption up by 67%, to approximately five billion gallons per year, according to Bloomberg Intelligence modelling published 27 April. Arizona is one of the fastest-growing data centre markets in the United States, yet the state depends heavily on Colorado River allocations that are already contested, overcommitted, and shrinking under a prolonged climate-driven drought. Water demand is projected to surge at exactly the moment long-running interstate allocation disputes remain unresolved and environmental stress on the basin intensifies.

For water utilities serving the US Southwest, this is an operational planning problem with no obvious resolution. Utilities are being asked to supply industrial-scale new loads for facilities that will be built quickly, with demand forecasts that change frequently and water availability that is declining in the long run. Community pushback is already a factor in development approvals, and several projects have faced delays after local opposition. For infrastructure planners and regulators, the lesson is that water impact assessments need to become as standard a condition of data centre development approval as energy connection and permitting.

Takeaway message: AI infrastructure needs water the sector does not have. Utilities that do not claim a seat at the planning table will have the decision made without them.

Thanks for reading! I hope you’ve enjoyed this week’s edition and look forward to seeing you next week!