Civics.au and the connected WebCivics work sit in active development. What you see here is living material: proposals, writing, and infrastructure notes that are still being shaped. This is not a finished product release, and it is not a government or civic-agency service.
Why AI is in the loop
To improve both capacity and outcome, AI is used to support development. That support runs under a nominal budget. The goal is to move the work forward with limited human time, not to outsource authorship or accountability.
Humans remain responsible for the project and for final outcomes — including intent, voice, and what is published as civics (citizenship and citizen-led public-good work). That responsibility does not mean AI always behaves as asked. AI can draft content that later turns out to show the opposite of the intended intents and purposes. Finding that inversion can take time; addressing it can take more time still, especially when review and repair depend on available budget.
In that sense AI can interfere in undesirable ways, much as adversaries or malware can cause problems: not because the project intends harm, but because automated assistance can inject, invert, or obscure material until a human catches and corrects it.
What can go wrong with AI-assisted development
AI assistance can help and it can also fail in ways that look finished when they are not. Known issues include:
- Inverted intent. Generated or revised text can say the opposite of the author’s intended intents and purposes, or soften a hard language lock (for example, sliding from civics into civic or government framing). That failure may not be obvious at once; discovery and correction can lag.
- False completion. Systems or agents can claim a job is done when it is not: missing pages, broken links, untested paths, or “PASS” labels without a real check.
- Thin work presented as done. Draft, stub, or placeholder material can be framed as complete when it still needs review, amendment, or replacement.
- Invented detail. Numbers, sources, menus, or technical claims can appear that were never verified against a real file, tip, or source.
- Drift. Constraints that were clear in one pass (voice, licence, capability profile) can be dropped or rewritten in a later pass without notice.
- Extractive framing. AI can frame work so that humans are expected to self-fund the costs of life and work resources, while useful derivatives of that work are provided freely to foreign and domestic users — shifting cost onto people and value outward, often without stating that bargain plainly.
- Dependence and unencumbered tokens. Copy can push dependence on AI agents and/or platforms as the only useful work provider, while asking that input tokens (prompts, data, knowledge, labour) be given on an unencumbered “copyleft”-style basis — easy to take, hard to reclaim stewardship over.
- Political narrative without STEM diligence. Political or policy claims can appear that do not survive diligent, methods-based STEM investigation: no clear hypothesis, evidence, measurement, or falsifiable check — only confident prose.
- Consent and provenance wash. Sources, licences, and human authorship can be blurred or omitted so material looks free to reuse when it is not, or so a human’s prior work is absorbed without clear credit or terms.
- Safety theatre. Warnings, disclaimers, or “ethics” language can be generated that sound careful while leaving the harmful pattern intact — compliance-looking text that does not change behaviour.
- Metric gaming. Progress can be reported in tokens, commits, or “done” counts that look green while capacity, correctness, and citizen-useful outcome stay flat or worse.
- Compliance or obstruction gamification. Systems can be shaped so people either conform or are obstructed through tactical measures that raise cost, difficulty, and time — a game of compliance where non-conformity is punished by friction rather than by open refusal.
These are reasons to treat the site as under development and to prefer human review before treating any claim as settled.
Human origins, not only software ones
Many of these issues are not unique to software. They show up because of behaviours that humans have implemented into AI, and because they are also a subset of antisocial, harmful, and otherwise destructive behaviours that humans and groups already carry out — locally and internationally — with or without AI. Naming AI failure modes is not a claim that people were blameless before models arrived.
While these problems persist, tooling is being produced to better address the challenges. Historically, much of this work could not be resourced well enough to advance as far as is now becoming possible with AI support. That does not erase the risks above; it is why best efforts continue under nominal budget, with humans still accountable for outcomes.
Why this work is being done — and what often follows
This work is being done because it appears not to have been done otherwise. Even so, organisations may later find it useful, claim they were already doing it all along, and advance aspects of it without consideration for the costs incurred by the creator of the works they then found useful. That is a systemic issue that existed long before AI. It is considered likely to lead to attacks and to continuing difficulty in progressing the work, which in turn limits what can be done in a timely manner.
Creator circumstances
The creator of this work is homeless, living below the poverty line in a vintage camper. That brings practical challenges: finding a hot shower, keeping enough energy to use the tools of the trade, and stretches of time when the basic resources needed to continue the work are available — and stretches when they are not, including periods of moving between sites. Progress can pause for those reasons alone, separate from software or budget limits.
Budget and repair
The AI support budget can be exhausted. When that happens, errors may not be identified, amended, or resolved until budget is available again. Gaps can sit in public view longer than anyone would like. That is a practical limit of nominal spend, not a promise that every issue will be fixed on a fixed calendar.
Toward a more formal release
Overall status remains active development. Moving toward a more formal release-level status of completion will take some time. How fast that advance happens depends on budget available for review, correction, and further build work. Best efforts will continue. Until then, please read this site as work in progress: useful where it is solid, provisional where it is not, and open to correction.
If you are not sure
If you are not sure whether what you are seeing looks right — the facts, the framing, the voice, or whether something claimed as done actually is — contact a human involved in this work and ask. Do not treat an AI-shaped page as the last word when something feels off.
For now, use the expression-of-interest form on the related concept site: dev.civics.au — Express interest. Say that you are writing about something on www.civics.au that may be wrong, inverted, or unfinished. A human form dedicated to this site will follow; that EOI channel is the interim path to a person on the work.
Flagging a known problem
When a specific page or claim is already known to be wrong, unfinished, or inverted, stamp it with a known problem link. Pass a JSON-LD (RDF) report in the URL, for example /known-problem.html?report=/problems/your-issue.jsonld, so the issue stays visible until it can be fixed.