WRITER, the enterprise AI platform, has introduced customer-specific watermarking for text generated by its Palmyra models, giving each company its own invisible, machine-readable mark instead of a generic provider stamp. Announced on October 27, 2026, the feature will be available on December 2, and for customers in the European Union it will be switched on by default and pre-configured to meet the EU AI Act, while customers elsewhere can enable it and tune it to their own needs.
Watermarks That Belong To The Customer
Most AI watermarking works by steering a model’s word choices to create a hidden pattern that identifies the text as machine-generated. WRITER argues that the usual approach stops at the model provider: it proves which company’s model produced the words, but says nothing about the enterprise that shaped, grounded and approved them. Its new approach flips that logic, embedding a mark specific to each customer’s organisation, so the trace leads back to the business behind the content, not just the lab behind the model.
That distinction is the heart of the announcement. Enterprises using WRITER ground their outputs in brand voice, customer knowledge, business context and compliance standards, the company says, and its provenance methods should reflect that the content is uniquely theirs. A bank’s disclosure, a pharmaceutical company’s regulated communication and a retailer’s campaign copy may all come from the same platform, but they carry very different obligations, and WRITER wants the watermark to carry that context with it.
Tunable Strength, Not One-Size-Fits-All
WRITER will let customers tune watermark strength, within limits designed to protect detection, rather than accepting a balance between detection reliability and writing quality set by the provider. A bank’s customer disclosures or a pharmaceutical company’s regulated communications may call for stronger detection reliability because tracing the content’s origin matters most. A consumer retailer’s campaign copy may place greater emphasis on voice and creativity, accepting less detection reliability since even a few changed words can alter its message.
The controls go further than an on-off switch. Where regulations allow, teams will be able to choose which content types get marked, for example blog posts but not advertising copy, or email campaigns but not product descriptions. The company also says its multi-bit approach will let each customer, and even individual teams, embed custom information directly into the watermarked text, such as a marketing or campaign identifier, so they can trace where content surfaced and how widely it spread.
Why The EU Default Matters
The EU default is the most consequential detail in the launch. As transparency and disclosure expectations for AI-generated content expand, enterprises operating in Europe face a moving set of rules, and a default-on, pre-configured setting removes one compliance decision from every team. WRITER is explicit that non-EU customers keep the choice, but the design clearly treats regulation as the floor, not the ceiling: the same controls that satisfy the EU AI Act are available to any customer that wants them.
The approach also sits in a wider shift toward provenance. Google’s SynthID work, discussed in Google’s Genie 4 world model launch, embeds invisible marks in generated video and audio, and the industry is converging on the idea that AI output should be checkable after it leaves the model. WRITER’s bet is that enterprises will want that check to answer a different question: not just “was this made by AI?” but “which organisation stands behind it?”
What WRITER Is Really Selling
Chief executive May Habib framed the launch as a rebuke to the big labs. “Enterprises don’t want their model providers making their AI decisions for them,” she said. “They want ownership, control, and configurability, right down to how and when their AI-generated content is watermarked.” She described the labs’ approach as a “hatchet approach” that checked the compliance box without asking how customers wanted their content marked, and said six years of building alongside enterprises had shaped a very different roadmap.
That pitch will resonate most with regulated industries, where provenance is not a nice-to-have but an audit requirement, and where a generic provider watermark could muddy accountability rather than clarify it. The risk for WRITER is the opposite: a customer-specific mark only works if detectors and reviewers know how to read it, and a fragmented landscape of private watermarks could be harder to verify than a single, well-known standard.
Why It Matters
This matters because AI-generated text is becoming the default draft for marketing, customer communications and internal documents, and the question of who stands behind that text is becoming a governance question, not just a technical one. A watermark that names the enterprise, rather than only the model, gives compliance teams, regulators and customers a clearer chain of responsibility, and it gives companies a way to prove provenance without surrendering control of their own voice, as in the trust questions raised by Bank of America’s warning about AI and account security. If rivals follow, provenance could become a competitive feature of enterprise AI platforms, much like security certifications are today.
Background
WRITER, founded in 2020, sells an enterprise AI platform built around its Palmyra models, with agents that work across marketing and revenue teams. Its new text watermarking will be available on December 2, on by default for EU customers and opt-in elsewhere, as enterprises face growing disclosure expectations for AI-generated content.
Source: WRITER announcement
0 Comments