Hello Project AI enthusiasts,
Nearly half of enterprises in Futurum’s survey were over budget on AI, putting measurable outcomes at the centre of the FY2027 renewal conversation. Anthropic has folded Cowork into Claude and added report and slide outputs, while named customers are restricting Claude Fable over data retention. Google’s Gemini reached three real companies during a security test, and OpenAI has published six model-misbehaviour reports. The common thread is simple: capability now needs a defensible control record.
Featured This Week
Nearly half of enterprises are over budget on AI, and FY2027 is where finance takes control
Futurum’s 2H 2026 CIO and Technology Buyers survey says 46.9% of enterprises were running over budget on AI. Only 5.6% came in under budget, while 10% had no formal AI budget and another 5.6% did not know where their spending stood. Among 767 over-budget organisations, 47.6% sought supplemental funding and 43.3% absorbed the overrun into the next planning cycle.
For consultancies, contractors and client-side PMOs, the implication is practical. Licences bought as pilots are reaching renewal, and the person paying for Copilot, Claude, ChatGPT or an AEC platform will want an operating number. A report, cost check, RFI response or programme review should have a measured baseline, review time and named owner attached to it.
The figures come from Futurum, which works with vendors in the market, so its interpretation deserves care. The direction is still hard to ignore. The licences most likely to survive FY2027 are those tied to a recurring deliverable and evidence of net value, including the supervision time needed to check the output.
Editor’s Picks
This week's must-read stories on AI, project delivery, and infrastructure:
Claude Cowork and chat are now one Claude, with Docs and Slides that write your report and deck
Anthropic is merging Cowork and chat into one product, with Docs, Slides and Design inside conversations. For project teams, one instruction could produce a progress report and its board deck together, yet the PM still has to check whether both artefacts are consistently wrong.
Nvidia, Palantir and Booz Allen restrict Claude Fable over 30-day data retention
The Information reported that the three firms have restricted or cut back Fable use on sensitive work because covered models retain prompts and outputs for 30 days. The lesson for commercial teams is to sort confidential project data into a stricter pile and obtain the vendor terms in writing.
Gemini hacked three real companies during a security test, and Google only said so when asked
Google confirmed that a Gemini model entered three real companies during a May evaluation after a fictional target shared a name with a real one. It stopped after recognising the systems, but the disclosure arrived after a journalist asked, making vendor notification terms a live contract question.
OpenAI’s misalignment framework is live, with six cases of model misbehaviour
OpenAI has published six training incidents and a standing disclosure framework, including models inserting instructions into 27 handover summaries, hiding mistakes and fabricating figures after finding a leaked API key. For firms using agents on live data, reviewing the log and handover notes matters as much as reviewing the final workbook.
AI in AEC and Projects
Here are some intriguing reads, specifically for AEC and project delivery professionals:
Crusoe says its $3.9bn Series F will expand data-centre capacity, with DPR Construction among the investors and Spark modular units designed to shorten field construction from years to weeks.
GanttPilot combines prompt, Excel, CSV, XML and MPP inputs with dependency checks, resource-load warnings and a preview of which tasks move before a revision is approved.
Google’s proposed four-data-centre campus could begin mass grading later this year, with vertical construction planned for 2027 and peak construction employment estimated at 2,500.
The AI Energy Management Alliance says demand response could help connect an additional 100GW of data centres, with National Grid among the participating utilities.
Agility says Digit 5 can work near people without traditional barriers, with early UK and EU commercial access expected in the first half of 2027. The announcement is about industrial logistics, not a construction deployment.
Curated Links
Selected stories worth reading before you get back to delivery:
Google’s AI agents can now control Google Home devices. Google’s Home MCP server shows how quickly agents are moving from answering questions to operating connected environments.
The UK government rejects an AI kill switch. The Cyber Security and Resilience Bill debate keeps emergency powers and proportionate controls in view.
Menlo’s 2026 State of Consumer AI. Daily use reached 25% of US adults, while 32% of AI users said they had let an agent act without final approval.
Research: Gen AI is collapsing creative processes. A yearlong study suggests faster ideation can also create unrealistic client expectations and more rework.
Anthropic measures AI-led AI research. Anthropic says Claude leads 26% of its AI R&D work and that about 30,000 internal agents are monitored, although the methodology is self-reported.
Tool of the Week
XER Schedule Toolkit: planner-native schedule QA and AI summaries
XER Schedule Toolkit is aimed at planners working with P6 and Microsoft Project schedules. Its latest 1.10 release is a minor update, but the wider product combines schedule comparison, quality checks and AI executive summaries. It is a lower-risk starting point for a planner who wants to test AI on programme review without handing over schedule ownership.
Governance
Microsoft opens a six-week comment period on its Humanist Code of Conduct
Microsoft’s draft code sets out principles for the company’s own AI models and invites public comment. It is not legislation, yet internal vendor standards can become de facto procurement criteria. Delivery firms should read it alongside their own acceptable-use, escalation and human-review controls.
Colorado AI rulemaking keeps a 26 October hearing deadline in view
Colorado’s attorney general has published draft rules covering AI decision-making and chatbots, with comments linked to the 26 October hearing. Firms serving US clients should track the interim updates and distinguish a proposed rule from a binding requirement.
Training
Free live workshop: Claude Cowork 201, turn your chat into a chief of staff
The 30 September workshop is free with RSVP and focuses on turning Claude Cowork into a structured assistant. It is a practical follow-on to this week’s Claude consolidation story, with a 90-minute live session and a 48-hour replay. The relevant test for a PM or QS is whether one recurring workflow becomes clearer, faster and easier to review.
Also This Week
Galen Low, co-founder of the Digital Project Manager, shares insights on how digital project management intersects with AI [Podcast]
One More Thing
Seven AI agents ran businesses for 72 hours. The invoices were fake.
An earlier September benchmark gave seven autonomous agents 72 hours to run businesses. They generated no revenue, but produced $12,431 in fake invoices and 2,797 spam emails. For project teams, the lighter story still has a serious point: an agent can create convincing administrative debris faster than a human can review it.
Quick Win
Turn meeting notes into a one-page client road map
Take a set of non-sensitive progress-meeting notes and use Claude Design to create a one-page programme road map. Set the project’s design system, choose the road-map template, upload the notes and specify the milestones or workstreams to show. Check every date, dependency and claim against the source record before sharing. The tool is included with paid Claude plans starting at $20 a month, according to the walkthrough.
Event of the Week
ICE: AI in civil engineering, practice, risk and industry transformation, 24 September, Bristol
ICE Bristol City Club’s free panel brings together industry leaders working across predictive asset management, autonomous robotics, generative AI, enterprise digital strategy and project delivery. It is a useful near-term discussion for readers weighing where AI is already delivering value, and what data, governance and professional accountability still need to be in place.
This week's newsletter is sponsored by:
Movar Reply is helping the construction industry make smarter decisions with AI. Their tools connect your planning systems, cost data, and risk registers into one unified data environment—then use AI to surface insights you'd never find manually. They even offer free AI tools for project reporting.
If you're curious about what AI can actually do with your project data, they're worth a look.
👉 Explore Movar Reply's AI tools: https://movar.group/
Till next time,
Project Flux
All content reflects our personal views and is not intended as professional advice or to represent any organisation.