The European Union now requires certain AI-generated content to be technically detectable. Claude is applying its response worldwide, including for Australian users.
While Australia does not currently require AI-generated content to be watermarked, the development raises a broader commercial question for Australian businesses: what happens when the use of AI becomes easier to detect?
What is Claude doing to AI-generated content?
Anthropic, the company behind Claude, is introducing machine-readable marking for AI-generated text and supported files in response to transparency requirements under the European Union’s Artificial Intelligence Act (EU AI Act).
They announced that new Claude models will mark AI-generated content in two ways:1
Invisible watermarks in AI-generated text: This is not a label that appears on the page. Rather, Claude builds a hidden statistical pattern into the words it generates. Detection technology can later look for that pattern.
Provenance information for supported files For supported files, Claude will also use C2PA provenance information. C2PA is an industry standard that allows signed information about a file's origin and history to travel with the file. In practical terms, it can provide a digital record showing that a file passed through Claude and whether that record has subsequently been altered.
That worldwide rollout is significant because the legal requirement that prompted Anthropic's announcement is Article 50 of the EU Artificial Intelligence Act. Anthropic says these measures will apply to supported Claude models worldwide, including when Claude is used outside Europe.2 Australian users therefore get the practical effect of an EU regulatory development even though Australian law has not imposed the same general requirement.
What does the EU AI Act require for AI-generated content?
Article 50 of the EU AI Act introduces transparency requirements for certain AI systems and AI-generated content.
Article 50 of the EU AI Act has been applied since 2 August 2026. AllAI providers must comply with Article 50 by 2 December 2026. Among other things, it requires providers of generative AI systems to ensure that certain AI-generated or manipulated text, images, audio and video are machine-readable and detectable as artificially generated or manipulated.3 The Act excludes AI that is only performing standard editing or does not substantially alter the material or its meaning.4
There are separate rules for businesses or organisations using an AI system in a professional capacity. They may have to visibly disclose the use of AI in deepfakes and certain AI-generated text published to inform the public about matters of public interest. The European Commission gives examples of public-interest subject matter including politics, public administration, justice and law enforcement, public health, consumer safety and significant economic, financial, scientific or cultural developments.
The Code of Practice is the practical implementation document. Section 1 deals with what AI providers can do to meet the marking and detection requirement. Section 2 deals with certain disclosure obligations applying to deployers. Signing the Code is voluntary; complying with Article 50 is not. The European Commission and AI Board have endorsed the Code as a recognised means of demonstrating compliance.5
Anthropic has signed the provider section of the Code.
Why did the EU introduce these rules?
The EU AI Act says that AI systems capable of generating content can create risks of impersonation and deception. It also recognises that AI can generate large quantities of synthetic material that is becoming harder for people to distinguish from authentic, human-created material. The risks identified include misinformation and manipulation at scale, fraud, impersonation and consumer deception.6 The EU's response is to make the origin of digital content easier to establish.
The introduction of the EU AI Act signifies that there is an emerging global view that AI should be regulated addressing public interest concerns of improper use of AI.
Is this just Claude?
Around 190 organisations had signed at least part of the EU Code by the end of July 2026. There are currently 82 signatories to the provider section, including Anthropic, Google, Meta, Microsoft and OpenAI.7 That does not mean all of those companies are putting the same invisible watermark into every piece of text they generate. The rules allow different technical approaches.
Why do Claude's AI watermarks matter for Australian businesses?
Australia is applying existing legislation to the way AI is developed and used. The Australian Government is encouraging greater transparency as a matter of good practice. The National AI Centre (the Australian Government's lead body supporting industry in adopting AI and part of the Department of Industry, Science and Resources) has published guidance on being clear about AI-generated content. That guidance is not legislation. It recommends that organisations consider whether disclosure is appropriate having regard to the context, the impact of the content and how much AI contributed to it.8
For Australian businesses, however, legal issue is whether the way AI is being used causes the business to contravene laws that already apply. The Australian Consumer Law, for example, does not need to mention ChatGPT, Claude or AI-generated images for it to apply to them.
The Australian Consumer Law already prohibits misleading or deceptive conduct and false or misleading representations about goods and services.9 AI creates new ways in which a misleading impression can be created. Using AI is not, by itself, misleading but what a business says (expressly or by implication) about the authorship, originality, expertise, authenticity or characteristics of what it is selling may be.
How could this play out in practice?
Use of AI | Commercial risk |
A property developer uses a photorealistic AI image | It matters whether the image is clearly presented as an artist's impression or instead gives the impression that it depicts the completed development as it actually exists. |
AI improves an ordinary product photograph | The key question is whether the resulting image still accurately represents the product the customer will receive. |
When should businesses disclose their use of AI?
There is no general Australian rule requiring a business to announce every occasion on which an employee used generative AI. Whether customers care is a different question. In many cases, they may not. A customer may be entirely comfortable knowing that AI helped a business format a document, summarise information or improve an internal workflow.
The reaction may be very different where the customer believes they are paying for human judgment, expertise, creativity or authenticity.
A person paying for specialist advice may feel differently if they discover that the substantive work was generated by AI. A business commissioning original artwork may care about AI use even if another customer buying an ordinary commodity would not. A client may care that its confidential information was entered into an AI system even though the final work itself is sound.
Watermarking does not answer whether any of those uses are acceptable. But it does weaken one assumption on which some AI use may currently depend: that nobody will know.
How do Australia’s privacy laws apply to AI and automated decision-making?
There is also a separate Australian transparency reform approaching under the Privacy Act 1988 (Cth).
From 10 December 2026, certain APP entities will need to include additional information in their privacy policies where they arrange for a computer program to use personal information to make decisions that could reasonably be expected to significantly affect an individual's rights or interests. That is an automated decision-making requirement, not a watermarking requirement, and it raises different issues.
We will look at those changes separately in What the new obligations surrounding automated decision-making mean for APP entities
What should Australian businesses do now?
AI has become an ordinary part of many workplaces and of the goods and services businesses provide. The next stage is likely to involve a balanced approach where AI disclosures are provided to offer individuals more visibility. For Australian businesses, the type of ‘balance’ you take have both legal and commercial consequences.
Existing laws concerning misleading conduct, contractual promises, privacy and other obligations still apply. At the same time, customers and counterparties will make their own judgments about how much AI involvement they consider acceptable.
Businesses should familiarise themselves with their current AI procedures and comfortably:
know where AI is actually being used. A policy drafted a year ago when generative AI’s capability were limited to just a chatbot, no longer reflect the breadth of tools available to businesses today.
decide where AI use matters. There is a meaningful difference between AI correcting grammar and AI producing the substance of work sold as specialist judgment or original creative output. Policies should recognise that difference;
confirm what has been promised. Contracts you may have entered into may impose obligations that go further than Australian legislation;
preserve origin where it may matter. If authenticity, authorship or compliance could later be disputed, keeping the original file and associated provenance information may be more useful than retaining only a screenshot or converted copy; and
consider whether there is an appropriate balance between preservation of reputation and compliance. Businesses should decide what role they want AI to play in their goods and services and whether that use is consistent with what their customers reasonably understand they are buying.
When should an Australian business disclose that it used AI?
There is no universal answer that applies for every situation. Businesses should consider whether they should take a principled approach for proactive disclosure or limit to what is legally necessary.
A better approach is to ask four questions:
Would the audience reasonably care that AI was involved? Using AI to correct spelling is very different from inventing a customer testimonial.
Does the content appear to depict something real? This is important for people, events, endorsements, product characteristics and experiences and the type of market you cater for. For example, if you have known dealings with artists, the active disclosure and reliance of AI may jeopardise commercial relationships.
Could someone make an important decision in reliance on it? The higher the consequence, the stronger the case for transparency and human checking.
Has the business promised anything about AI use? Check contracts, tender conditions, client policies, internal procedures and industry-specific rules.
If the answer to one or more of those questions is yes, the business should at least consider whether an AI disclosure, visible label, internal record or additional human approval is appropriate.
If your business is reviewing how it uses AI, including disclosure, customer communications, privacy or contractual obligations, contact Julian Courtney-Stubbs or Sandra Bejo to discuss how these developments may affect your organisation and the practical steps you should take now.
1 https://support.claude.com/en/articles/16266773-how-claude-marks-ai-generated-content
2 Ibid.
3 Regulation (EU) 2024/1689 (Artificial Intelligence Act), Article 50(2).
4 Ibid.
5 European Commission, Code of Practice on Transparency of AI-generated Content.
6 Regulation (EU) 2024/1689, recitals 132–133.
7 https://digital-strategy.ec.europa.eu/en/policies/code-practice-ai-generated-content
8 https://www.ai.gov.au/staying-safe-and-responsible/essential-ai-practices/be-clear-about-ai-use
9 ACL
This publication has been prepared for general guidance on matters of interest only and does not constitute professional legal advice. You should not act upon the information contained in this publication without obtaining specific professional legal advice. No representation or warranty (express or implied) is given as to the accuracy or completeness of the information contained in this publication and to the extent permitted by law, Cowell Clarke does not accept or assume any liability, responsibility or duty of care for any consequences of you or anyone else acting or refraining to act in relation on the information contained in this publication or for any decision based on it.