A routine social media post by a small New Zealand council has ballooned into an international cautionary tale, exposing the perilous gap between the promise of generative AI and the reality of public accountability. On May 28, the Waimakariri District Council republished a short video announcing service changes for the King’s Birthday Weekend—a public holiday honoring the monarch’s official birthday. But rather than a straightforward bulletin, residents were greeted by a jarring, machine-crafted clip that many quickly labeled “AI slop.” The fallout has since forced a reckoning over how local governments deploy tools like Microsoft Copilot without clear oversight, training, or even a basic framework for accountability.
A Holiday Message Gone Wrong
The 45-second video, posted on the council’s official Facebook page, was intended to remind residents of altered rubbish collection schedules, library closures, and other administrative tweaks for the long weekend. Instead, its uncanny valley aesthetic—a collage of synthetic imagery, a monotone text-to-speech narration, and a script riddled with awkward phrasing—drew instant ridicule. One Facebook user commented, “This looks like it was made by a robot that’s never met a human.” Others lambasted the council for using taxpayer-funded tools to produce what they saw as a lazy, disrespectful communication.
Within hours, the post had accumulated hundreds of reactions, most expressing anger or amused contempt. Local media picked up the story quickly, and the term “AI slop” trended in New Zealand’s digital circles. The council initially defended the video as a creative, cost-effective way to modernize its public outreach, but as the backlash intensified, it deleted the post and issued an apology. “We underestimated the importance of a human touch in our communications,” a council spokesperson later admitted. “We will review our processes to ensure this doesn’t happen again.”
The Anatomy of AI Slop
“AI slop” has become shorthand for the growing deluge of low-effort, algorithmically generated content that clogs social feeds, search results, and now apparently municipal channels. Coined by developer and AI commentator Simon Willison, the term originally described the flood of bizarre, often grotesque AI images on platforms like Facebook. But it has since expanded to encompass any automated output that prioritizes speed and volume over quality, accuracy, or human relevance.
The Waimakariri video ticked every box. The visuals appeared to be stock-like imagery plausibly generated by an early version of Microsoft’s Designer tool or a similar diffusion model—people with awkward, asymmetrical faces, surreal backgrounds, and a general plasticky smoothness. The voiceover, likely produced by a neural text-to-speech engine, stumbled over Te Reo Māori pronunciations (the council’s district includes significant Māori populations) and delivered its lines with the emotional range of a GPS navigator. Most damningly, the script itself contained factual errors about the holiday’s history, mistakenly conflating the King’s Birthday with the actual birthday of the monarch—a slip that royal-watching Kiwis found disrespectful.
Microsoft Copilot and the Public Sector
While the council never explicitly confirmed which AI tools were used, multiple clues point toward the Microsoft 365 ecosystem. Waimakariri, like many New Zealand councils, is a Microsoft shop, using Outlook, Teams, and SharePoint. Microsoft Copilot—a suite of generative AI assistants embedded in these apps—has been aggressively marketed to government agencies as a productivity booster. One Copilot feature, the ability to generate video storyboards and narrations from a text prompt, aligns suspiciously well with the output seen in the debacle.
Microsoft’s Copilot for Microsoft 365 integrates with Word, PowerPoint, and Clipchamp, a relatively new video editor that includes AI-powered voiceover and stock media generation. A council staffer seeking to quickly drum up a holiday notice could theoretically input a few bullet points into Copilot, ask it to expand them into a script, and then auto-generate a video complete with AI voice and visuals—all without leaving the Microsoft environment. That frictionless pipeline, while a marvel of engineering, becomes a liability when the humans in the loop are not trained to scrutinize the output. As one New Zealand IT consultant noted on Twitter, “This isn’t an AI failure; it’s a workflow failure. Someone hit ‘publish’ without putting their brain in gear.”
Accountability in the Age of Automation
The incident has stirred a deeper debate: when a public body disseminates AI-generated information that is inaccurate, offensive, or simply embarrassing, who is held responsible? The Waimakariri Council’s initial reaction—a defensive posture followed by a generic promise to “review processes”—is emblematic of an accountability vacuum. There is no obvious individual to blame, no mechanism for public redress, and no clear standard for what constitutes acceptable use of generative AI in government communications.
Under New Zealand’s current digital government strategy, agencies are encouraged to experiment with emerging technologies, but there are no binding rules on AI output quality or mandatory human oversight. The Algorithm Charter for Aotearoa New Zealand, adopted in 2020, asks signatories to commit to transparency, fairness, and human accountability, but it is voluntary and lacks enforcement teeth. Waimakariri is not a signatory.
Legal experts point out that any council content, regardless of its origin, constitutes official advice. If an AI-generated post misstates rubbish collection times, and a resident relies on that to their detriment, the council could face liability under consumer protection laws or even negligence claims. The fact that a machine wrote the error does not absolve the authority. “AI is not a shield,” says Jenny Martin, a Wellington-based public law specialist. “The council remains vicariously liable for its outputs, just as it would be if an intern wrote a wrong press release.”
A Broader Pattern of Municipal AI Blunders
Waimakariri is hardly alone. Around the world, well-intentioned local governments have stumbled into AI-generated embarrassments. In January 2024, the City of San Francisco was criticized after its official chatbot gave out bogus tax advice. In Australia, the Victorian Department of Education used AI to draft a letter about school closures that contained fabricated statistics. The common thread is a rush to adopt flashy new tools without commensurate investment in training, auditing, or ethical guardrails.
These missteps are not merely cosmetic. They erode public trust when institutions already face skepticism. A 2023 survey by the OECD found that only 38% of New Zealanders trust their central government, and trust in local councils is often even lower. Churning out robotic, error-prone content reinforces the perception that bureaucracies are faceless and incompetent.
The Technical and Cultural Fix
So what would a responsible AI usage framework for local councils look like? Experts suggest several key elements. First, mandating a “human-in-the-loop” checkpoint for all public-facing content. Every AI-generated draft must be reviewed by a trained communications officer who understands both the technology’s limitations and the community’s expectations. Second, implementing clear labeling: if AI played a role in creating a public message, the public should know. Third, establishing incident-reporting protocols so that when AI does go wrong—as it inevitably will—the response is swift, transparent, and focused on learning rather than blame deflection.
Microsoft itself provides tools that could help. The company’s Responsible AI dashboard, available to enterprise customers, allows organizations to detect statistical anomalies, fairness issues, and factual errors in model outputs before deployment. Yet councils rarely have the data science expertise to use such tools effectively. That skills gap is a significant barrier.
There is also a cultural component. Council leaders must move beyond the allure of “AI washing”—slapping smart technology onto inefficient processes and calling it innovation. True digital transformation requires rethinking workflows, not just automating existing ones. If a human would be ashamed to publish a script or video, no machine should be allowed to do it on their behalf.
The Future of Local Government AI
Despite the uproar, the Waimakariri incident may ultimately do more good than harm if it provokes a long-overdue conversation. Already, the New Zealand Society of Local Government Managers has announced a working group to draft guidelines for generative AI use in councils. Alistair McDonald, the chief executive of the Canterbury-based Selwyn District Council and a board member of the society, wrote in a LinkedIn post that “this was a preventable fail, and it’s on all of us to ensure our teams are equipped to use these powerful tools responsibly.”
For Microsoft, the episode is an uncomfortable reminder that its Copilot suite—while immensely capable—is being deployed in contexts far removed from the controlled enterprise settings it was designed for. The company has since added more prominent disclaimers to Copilot’s video generation feature warning users to verify content, but technical fixes alone will not solve the governance problem. As the line between human and machine creativity blurs, citizens will increasingly demand to know who—or what—is talking to them.
Back in North Canterbury, the council has promised to rebuild trust one post at a time. Its next Facebook update, a straightforward text-and-photo announcement about pothole repairs, was overwhelmingly well-received. Sometimes, it seems, the best AI strategy is knowing when not to use it at all.