WRITER lets enterprises watermark AI text with their own key, not the model's

Watermarks tied to a company's private key

According to WRITER, text generated by its Palmyra models will carry an invisible, statistical signal tied to the customer's private key once watermarking is enabled. An accompanying detector looks for that signal and decodes the information it contains. The company argues that most watermarking points back to the model provider, even though a finished piece of content also reflects a company's brand guidelines, customer insights, compliance rules and editorial judgment.

Watermarking is also becoming a regulatory topic. WRITER notes that laws such as the EU AI Act require certain AI providers to watermark AI-generated content, including text.

The company says any text watermark can affect word choice and quality, and that a stronger watermark means a greater impact. WRITER lets customers tune the strength to find a balance that suits them. Enterprises can also choose which content types are watermarked, for example blog posts but not ad copy, or email campaigns but not product descriptions. Teams can embed custom information in the watermarked text, such as a campaign ID, to trace the reach of their content.

Defaults, availability and third-party models

WRITER says it will enable text watermarking by default for EU enterprises to support compliance with the EU AI Act. Outside the EU, customers can choose to turn it on to protect their IP, track content or meet other needs. Text watermarking will be available by December 2.

The company adds that if customers use third-party models in WRITER, the watermark policy of that model provider applies.

Approvals, admin controls and security

For agentic workflows, WRITER introduced Playbook approvals, which add built-in handoffs for human review. Instead of a workflow pausing until someone remembers to resume it, an approval creates a structured checkpoint, and the work moves forward only when the right person signs off. New Playbook admin controls determine who can create Playbooks and who can publish them to shared libraries.

On security, WRITER cites recent research showing that user-generated and imported skills can hide malicious instructions that lead agents to steal data, delete files or carry out ransomware-style attacks. It shipped a skill security scanner that reviews every skill. When the company turned it on internally, it disabled a few skills its teams relied on, and some of those flags were false positives. Customers can therefore choose whether flagged skills are disabled automatically.

A new data loss prevention guardrail blocks sensitive information such as PII, financial data and credentials from reaching models and specified teams. Customers who use third-party guardrails can keep doing so.

MCP access, live dashboards and agent steering

WRITER MCP lets users run Playbooks, fetch work and pull in WRITER Meet notes from any MCP-compatible client. It works with existing WRITER seats and permissions. According to the company, context from Enterprise Brain will come to WRITER MCP next month.

Dashboards and microsites can now pull live data from connectors and other data sources, so shared views stay current. Users can customize the theme and, in edit mode, select parts of a dashboard and ask WRITER Agent for a targeted update without regenerating the whole deliverable. Updates are staged in drafts, and publishing pushes changes to the live deliverable without resharing the link.

Users can also send a message while WRITER Agent is working to change its course, and queue follow-up requests that can be reordered as priorities change.

What it means for enterprise customers

WRITER frames the release around giving companies control over how AI-generated content is identified and how agent workflows are governed. For EU customers, watermarking will be on by default from December 2, while others can opt in. The tunable strength and per-content-type settings let teams weigh traceability against the effect on wording. The company says this control is what moves AI use from siloed experiments to coordinated work.