Signal file: where is AI governance falling behind?
From AI-driven layoffs and shadow AI risk to mounting infrastructure costs and office mandates, this week's signals examine who is accountable when technology shapes decisions about work
Technology is increasingly making, or at least influencing, decisions that used to sit squarely with people. This week’s signals point to organisations and regulators grappling with where accountability actually sits, whether that is a chatbot that gives the wrong answer, an algorithm used to select employees for layoffs, or the spread of unauthorised AI tools inside a company’s own systems. Together, the developments suggest a workplace landscape where the pace of technology adoption is outstripping the governance built to manage it.
Who answers when AI gets it wrong
New reporting from the Washington Post examines how the rapid spread of chatbots and AI agents is intensifying a decades-old question from science fiction: who should be held responsible when an artificial intelligence causes harm. As companies hand agents an expanding range of tasks, courts, employers and policymakers are increasingly being forced to work out where liability sits when something goes wrong. The debate touches everything from customer service bots to internal decision-making tools, and comes as organisations race to deploy agents faster than governance frameworks can keep pace.
In action: Define accountability before deployment. Organisations introducing AI agents will need clear internal ownership for outcomes, alongside technical safeguards, if they want a defensible position when something goes wrong.
AI-driven layoffs land in court
A group of 26 current and former Meta employees has filed a lawsuit in federal court alleging the company used internal AI systems, including performance ratings, activity-monitoring data and algorithmic rankings, to select workers for its recent 8,000-person layoff round. The suit claims those scoring systems disproportionately affected employees on protected medical, parental or disability leave, since their measured output could not accumulate in the same way as colleagues who were working. Meta has denied the claims, saying workforce decisions were made by people, not AI.
In action: Stress-test workforce algorithms against protected categories before they are used. Any tool that scores or ranks employees for redundancy needs an individualised review process that accounts for leave and accommodations, not just aggregate output data.
Employees quietly widen their own risk exposure
New research from WatchGuard Technologies finds that 64% of employees admit to using unauthorised AI tools for work, contributing to a shadow AI problem that most organisations lack the visibility to manage. The same survey found that a majority of employees continue to reuse passwords, use public wifi for work, and access corporate systems without VPN protection, compounding the risk. The findings point to a widening gap between how employees actually work and the controls organisations believe they have in place.
In action: Close the visibility gap before writing the policy. Organisations need a clear picture of which AI tools are actually in use across the business, not just a list of approved ones, if shadow AI risk is to be properly managed.
The cost of the AI build-out reaches the payroll
New reporting from HR Brew details how Oracle’s restructuring costs, including severance and contract terminations, nearly quadrupled to $1.8 billion in its last fiscal year, as the company cut 21,000 jobs while increasing capital expenditure on AI data centre infrastructure by 162%. Analysts note the layoffs are not purely an AI story: Oracle is also still absorbing an underperforming $28 billion healthcare acquisition, and the cuts look as much like an attempt to close that value gap as AI-driven efficiency. The findings are a reminder that AI investment and workforce reduction are increasingly presented as a single narrative, even when the underlying causes are more mixed.
In action: Separate the AI narrative from the underlying business case. Leaders attributing restructuring to AI should be able to show their working, since conflating unrelated cost pressures with technology efficiency can undermine trust with employees and investors alike.
California doubles down on office attendance
California governor Gavin Newsom has doubled the state’s in-office requirement for around 90,000 state workers from two days a week to four, according to Bloomberg, setting up a clash with SEIU Local 1000 over telework as a bargaining right. The union has filed an unfair labour practice complaint alleging the administration failed to negotiate the policy in good faith, while workers have raised concerns about commuting costs and insufficient office space to accommodate the return. Newsom has argued the mandate will improve collaboration and support downtown businesses.
In action: Treat large-scale return-to-office mandates as a workplace planning exercise, not just a policy announcement. Space, seating and commuting support all need to be resolved before the deadline, not worked out afterwards.
