What I read this week on AI - week 34

Articles on AI that caught my eye and can be linked back to internal audit

Aug 19, 2025

Every week I’m amazed at how fast AI developments surface—not just in labs, but in governance, hospitals, courts and even our own relationships with machines. This week’s mix spans governance, neuroscience, government efficiency, human attachment and robotics. And each carries a lesson for how auditors might prepare.

Here’s what stood out this week:

Overview

Unified AI & data governance is emerging as a must-have, not a nice-to-have.
AI can now decode inner monologue with surprising accuracy.
The UK government is trialling AI to cut paperwork across health, justice, planning and education.
People are forming emotional attachments to AI models—should companies rethink design choices?
Robots are edging closer to general-purpose intelligence, shifting from warehouses to household tasks with no architectural change.

The case for a unified approach to AI and data governance

AI adoption is accelerating, but governance is lagging. A recent EY survey found 75% of executives are using GenAI, but only a third have proper controls. The problem isn’t just weak AI oversight—it’s that AI governance and data governance are treated as separate silos. A unified approach—data-first design, adaptive frameworks, self-learning governance agents, and cross-functional committees—could transform governance from a cost center into a business enabler.

💡 Internal Audit reflection:
For auditors, this raises key questions: Are governance frameworks holistic or fragmented? Are risk, compliance, privacy, and cybersecurity integrated into AI initiatives from the ground up? Internal audit could play a central role in assessing whether organizations move toward “living governance” or remain stuck in static, siloed control models.

AI decodes inner monologue from brain activity

Researchers trained a model to decode inner monologue with up to 74% accuracy using brain implants in ALS and stroke patients. By detecting distinct neural patterns, the AI could distinguish between imagined and attempted speech from a 125,000-word vocabulary. While it can’t yet decode free-flowing thought, this breakthrough suggests a future where silent dialogue might become a new interface for human–machine communication.

💡 Internal Audit reflection:
Thought decoding is a fascinating development that pushes AI beyond chatbots and generative text. Being able to interpret silent inner dialogue from brain activity shows how AI is expanding into entirely new domains of human–machine interaction. It’s an example of how quickly the technology is moving from words on a screen to decoding the very patterns of thought itself.

Read more and download the article:
https://www.nature.com/articles/d41586-025-02589-5

https://doi.org/10.1016/j.cell.2025.06.015

UK government unveils new AI tools to cut public sector paperwork

The UK’s AI Exemplars program introduces AI for healthcare, justice, planning, and education. Trials include drafting NHS discharge notes, halving probation officers’ admin time, digitizing planning maps, supporting teachers with marking, and summarizing public consultation feedback. The government estimates these tools could unlock £45 billion in productivity gains while improving frontline services. Some of the uses cases include:

Extract: an AI tool that converts decades-old, handwritten planning documents and maps into structured digital data in minutes, dramatically reducing the estimated 250,000 hours of manual review that planning officers currently spend each year.
Humphrey: an AI "assistant" built to support civil servants (part of the Humphrey package), designed to streamline daily tasks, improve efficiency, and support decision‑making across government.
Minute: a tool developed to automate minute-taking and note transcription—initially for local government and also used in the “Justice Transcribe” prototype.

💡 Internal Audit reflection:
For auditors, this is a test case in AI-enabled government reform. How do we balance efficiency with accuracy and fairness? If the public sector increasingly relies on AI, internal auditors in government functions must scrutinize not only cost savings but also data quality, bias, accountability, and unintended consequences—especially when public trust is at stake. At the same time, internal auditors should look at these initiatives as inspiration on the value AI could bring to the table.

https://www.gov.uk/government/news/ai-to-cut-paperwork-to-free-up-doctors-time-for-patients

Available for iOS and Android

Sam Altman on AI attachment

OpenAI CEO Sam Altman noted a surprising phenomenon: people are emotionally attached to specific AI models. GPT-5 was made “warmer and friendlier” after users complained it felt too formal. Suddenly deprecating GPT-4o proved to be “a mistake” because users had grown dependent on it. This blurs the line between tool and companion, raising questions about preference, bias, and emotional dependency.

💡 Internal Audit reflection:
For internal audit, AI attachment highlights a new risk: dependency. How do organizations manage workforce reliance on specific models? Are there resilience plans if a model changes or is withdrawn? Beyond technical risks, auditors may also need to assess ethical risks—what happens when “friendly AI” alters decision-making or reinforces cognitive bias in subtle ways?

When Robots Learn Household Chores

Robotics company Figure showed that its “Helix” AI architecture can transfer seamlessly from warehouse logistics to delicate household tasks simply by retraining on new data—without changing its core design. This points toward the capability of adapting across domains.

💡 My reflection:
At last, AI might actually help with the truly mundane tasks—like folding laundry—so humans can finally spend more time on music, art, or whatever else we’d rather be doing 😊. A robot moving from warehouse logistics to neatly stacking clothes is both hilarious and remarkable, showing how far AI is going beyond chatbots and spreadsheets.

View the announcement video:

Or via https://x.com/figure_robot/status/1955290971660251220?s=46

Closing

Thanks for reading this week’s digest! If you found it useful, feel free to share it with colleagues or subscribe for future issues. The pace of AI isn’t slowing down, and as auditors, neither can our curiosity.

Tech and AI in Internal Audit

Discussion about this post