A second AI Act deadline is coming, and most communications teams have already stopped tracking it

A second, narrower deadline inside Article 50 falls on 2nd December 2026, and it sits on a different desk from the first.

Michael Reaney  ·  30 October 2026  ·  4 min read

Article 50 of the EU AI Act, the provision requiring deepfake images, audio and video, and AI-generated text published on matters of public interest, to be disclosed, became enforceable on 2nd August 2026. Communications and legal teams who spent the summer working out whether a ghostwritten executive post needed a label (it does not, provided a named person with the authority to change or reject it genuinely reviewed it) may well have filed that deadline as handled.

A second, narrower deadline inside the same article falls on 2nd December 2026. Law firms tracking the rule, among them Baker Botts and Ropes & Gray, describe the same shape: the Digital Omnibus on AI, in force since 27th July, delayed only the machine-readable marking duty set out in Article 50(2), the technical requirement that the provider of a generative system ensure its output carries a detectable, machine-readable signal. Every other part of Article 50, including the disclosure duty that falls on whoever deploys the content, has applied since August and was not touched by the Omnibus. The marking delay is a grace period for systems already on the market or in service before 2nd August 2026; a system placed on the market after that date had no grace period and should already comply.

The distinction matters because the two duties sit on different desks. The disclosure obligation is an editorial decision: for public-interest text, did a named person genuinely review this before it went out; for a deepfake video, voice or image, was it labelled. The marking obligation is a vendor and procurement question: does the generative tool itself, whatever is producing a company's AI-assisted drafts, images or video, write a machine-readable signal into its output, and can the vendor show that in writing. A function that closed the first question in August can still carry risk on the second. The legal duty sits with the vendor, but the output goes out under the company's name, and a company that has built its own generative tool, even on top of a commercial model, may count as a provider itself. Machine-readable marking is also unlikely to have been on the checklist when most tools were approved for procurement.

Three checks worth running against a communications or marketing function before 2nd December

1. List every generative AI tool in the content pipeline, from drafting through to image and video production, and note whether each one was on the market or in service before 2nd August 2026. Only those have until 2nd December; anything newer should already comply.
2. For each one, get written confirmation from the vendor that its output carries Article 50(2) machine-readable marking, not only a visible disclosure line.
3. Check that the team understands the marking duty and the disclosure duty are separate. A tool that marks its output correctly does not discharge the organisation's own disclosure duty. For public-interest text, a genuine named human review closes the disclosure question; for deepfake video, audio and images, review does not remove the need for a label.

Worth reading

• Baker Botts, "EU AI Act Article 50 Transparency Obligations Go Live" (law firm client briefing, published 10th September 2026). Confirms Article 50's transparency obligations became enforceable across the EU on 2nd August 2026, sets out what providers and deployers must each do, and dates the marking grace period to 2nd December 2026 for generative systems already on the market. Baker Botts, 10th September 2026
• Ropes & Gray, "You Talkin' to Me? Operationalising the EU AI Act's Transparency Obligations" (law firm client briefing by Edward Machin, published 3rd August 2026). Walks through Article 50 as four separate obligations and confirms the Digital Omnibus touched only one of them, the Article 50(2) marking duty, and then only for systems already on the market or in service. Ropes & Gray, August 2026
• WTW, Reputational Risk Readiness Report 2026 (survey of 500 senior executives responsible for risk strategy, across five sectors and 20 countries, fieldwork with Coleman Parkes Research in January and February 2026, published April 2026). Only 37 per cent of respondents now say they understand the key hotspots of negative sentiment affecting their brand, down sharply from 56 per cent in the 2024 edition, even as 82 per cent place reputation among their top three or top five risks and 56 per cent report a low appetite for reputational risk, up from 36 per cent two years ago. The share with a formal crisis team also fell, from 87 to 80 per cent. The shape echoes the one logged in the last two editions for the CCO's remit (more mandate, no more resource to meet it): the stakes placed on reputation are rising while the means to read and manage it are thinning. WTW, April 2026

The moves

• Michael Baker International, the Pittsburgh-based engineering and consulting firm, announced the appointment of Chris Phillips as chief communications officer on 6th October 2026. She joined from Bentley Systems, where she was head of external communications, with earlier roles at E-Space, Viasat, Entropic Communications and Ruder Finn/Finn Partners across more than 25 years in corporate communications. A smaller-profile move than the CCO transitions logged in previous editions, included for completeness in a thin fortnight for named appointments. Reported by O'Dwyer's PR News, 7th October 2026. O'Dwyer's PR News, 7th October 2026