On 2 August 2026, the European Commission's AI Office and national authorities began enforcing the Artificial Intelligence Act. The high-risk regime, originally due the same day, moved to 2 December 2027 for the Annex III categories and 2 August 2028 for AI embedded in physical products, under the AI Omnibus agreed politically in May 2026 and in force since 27 July.
That inversion matters for pharma because the labelling duty mentioned in the article 50 is related to published content, and the Commission's list of qualifying subjects includes public health information.
What did the AI Act actually make enforceable on 2 August 2026?
Four blocks: general-purpose of AI models, the prohibited practices, the transparency rules of Article 50, and AI literacy. Conformity assessment, technical documentation and post-market monitoring for high-risk systems are not among them.
According to the Commission's press release of 31 July 2026, interactive systems must now tell users they are dealing with AI rather than a person ( customer service chatbot, Healthcare chatbot, phone voice assistant, AI-generated email..), deep fakes must be labelled, and AI-generated or altered content must carry machine-readable marks so it can be detected. More than 180 organisations have signed the Code of Practice on transparency of AI-generated content. The Omnibus also allows the processing of special categories of personal data to detect and correct bias.
What does AI Act Article 50 require pharma to label?
Two things, and only once published: deep fakes, and AI-generated or manipulated text published to inform the public on matters of public interest.
The Commission's FAQ on Article 50, updated 24 July 2026, splits the duties.
Providers must inform people interacting with their system and mark generative outputs in machine-readable form. Deployers, defined by the Commission as bodies using an AI system under their authority, must inform people exposed to emotion recognition or biometric categorisation systems, and carry the labelling duty above.
The Commission's list of qualifying subjects includes public health, along with scientific developments that may be a relevant subject of public debate. It does not name specific content types.
What counts as human review under Article 50?
Substantive examination by someone with relevant professional judgement. Text that has undergone human review or editorial control does not need labelling, where a person holds editorial responsibility for the publication. The Commission states that superficial or purely procedural checks such as spell-checking do not qualify.
Which Article 50 dates and penalties apply?
Content generated before 2 August 2026 needs no retroactive labelling. For systems placed on the market before that date, the marking obligation of Article 50 applies only from 2 December 2026, the nearest fixed date in this file. Fines reach 15 million euros or 3 percent of worldwide turnover, with proportionality available for SMEs and small mid-cap companies.
How is the United States approaching clinical AI evaluation?
Through a one-month expert sprint aimed at a consensus set of principles, convened by the White House Office of Science and Technology Policy, the FDA and the Office of the National Coordinator for Health Information Technology.
STAT reported on 23 July 2026 that the sprint is split into a written phase and a discussion phase, and covers the benchmarking and evaluation of clinical AI. The same reporter documented, on 5 August 2026, a clinical AI demo day held on 8 July at the FDA's White Oak headquarters, where FDA and CMS officials hosted leaders from ten companies including Anthropic, Microsoft AI and Amazon One Medical. According to STAT, the meetings had not been publicly announced, and the event offered attendees an opportunity to shape policy as officials wrestle with regulating health AI and with how to pay for it.
Does the AI Act apply to AI used in drug discovery?
Article 2 provides that it does not apply to AI systems or models developed and put into service for the sole purpose of scientific research and development, nor to research, testing and development activity prior to placing on the market, with testing in real-world conditions excluded from that carve-out. The Annex III areas the Commission lists are biometrics, critical infrastructure, education, employment, migration, asylum and border control.
The Novo Nordisk agreement of 10 August 2026 shows what is being deployed meanwhile. According to Fierce Biotech, it gives Novo access to Amazon Bio Discovery, which offers models trained on biological datasets, and will see it deploy Amazon Bedrock AgentCore to run AI agents across its operations, with tools intended to identify targets, design therapies and bring together clinical, genomic and imaging data. Novo said the partnership had already reduced clinical documentation time and made 25,000 employees more productive, so those results predate the new agreement.
What should pharma teams do before December 2027?
What follows is our analysis rather than reporting.
- Audit published output against Article 50. The boundary the Commission draws runs through editorial process rather than technology. Which published outputs were generated or materially modified by a model, and which of those passed substantive human review with named editorial responsibility, is answerable now.
- Classify systems against Annex III, Annex I and the research exclusion. That classification decides whether either postponed deadline applies at all.
- Keep a human in the loop, and know where every figure comes from. The Article 50 exemption turns on substantive human review with named editorial responsibility. At Tekkare we use AI to build custom intelligence reports on top of reliable public data sources, with human review before anything is issued, and every figure stays linked to the source it came from.
Sources
- European Commission, DG CONNECT. "AI Omnibus enters into force." News article, 27 July 2026. https://digital-strategy.ec.europa.eu/en/news/ai-omnibus-enters-force
- European Commission, DG CONNECT. "Commission starts enforcing AI Act rules and new transparency requirements on 2 August." Press release, 31 July 2026. https://digital-strategy.ec.europa.eu/en/news/commission-starts-enforcing-ai-act-rules-and-new-transparency-requirements-2-august
- European Commission, DG CONNECT. "Transparency obligations under Article 50 of the AI Act." FAQ, last updated 24 July 2026. https://digital-strategy.ec.europa.eu/en/faqs/transparency-obligations-under-article-50-ai-act
- European Commission, AI Act Service Desk. "Timeline for the Implementation of the EU AI Act." Accessed 17 August 2026. https://ai-act-service-desk.ec.europa.eu/en/ai-act/timeline/timeline-implementation-eu-ai-act
- European Commission, AI Act Service Desk. "Article 2: Scope." Accessed 17 August 2026. https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-2
- Aguilar, Mario. "HHS to convene experts on standards for clinical AI." STAT Health Tech, 23 July 2026. https://www.statnews.com/2026/07/23/hhs-convenes-experts-on-clinical-ai-health-tech/
- Aguilar, Mario. "Federal regulators invite industry, researchers, and lobbyists to closed-door meetings on clinical AI." STAT, 5 August 2026. https://www.statnews.com/2026/08/05/federal-regulators-invite-industry-closed-door-meetings-clinical-ai/
- Maddox, Will. "Novo Nordisk and Amazon sign AI drug discovery pact, launch London innovation hub." Fierce Biotech, 10 August 2026. https://www.fiercebiotech.com/biotech/novo-nordisk-and-aws-sign-ai-drug-discovery-pact-launch-london-innovation-hub
