Skip to content
HACKTECH
RegulationReported Published 7 sources

EU AI Act enforcement and transparency rules start as AI Omnibus delays high-risk rules

The EU AI Act became applicable on 2 August 2026, when the European Commission's AI Office and national authorities began enforcing it and its transparency rules for chatbots, deepfakes and AI-generated content took effect, according to the Commission and Help Net Security. The AI Omnibus amendments, proposed on 19 November 2025, agreed on 7 May 2026 and in force since 27 July 2026, set 2 December 2027 for Annex III high-risk systems and 2 August 2028 for high-risk AI embedded in regulated products. Help Net Security reported that ignoring the transparency obligations risks fines of up to €15 million or 3% of worldwide annual turnover, whichever is higher.

  • EU AI Act
  • AI Office
  • AI Omnibus
  • AI transparency
  • general-purpose AI
  • deepfakes

What the sources report

  • From 2 August 2026 the AI Office holds enforcement powers over general-purpose AI models: it can request technical documentation, evaluate models, require corrective measures and issue fines (European Commission).
  • Under the transparency rules, chatbots must identify themselves as automated, deepfakes need a label and machine-made or edited content must carry machine-readable marks. The Commission published transparency guidelines on 20 July 2026, and Help Net Security reported that the Commission released a first list of over 180 organisations that signed the voluntary Code of Practice on transparency of AI-generated content.
  • High-risk obligations, which now apply from 2 December 2027 for Annex III systems and from 2 August 2028 for AI embedded in regulated products, include logging of activity for traceability and a high level of accuracy, resilience and cybersecurity. Providers and deployers will also report serious incidents and malfunctioning.
  • According to Help Net Security, the Commission said providers of the most advanced general-purpose AI models must address risks including cyber offence, loss of control and harmful manipulation.
  • From 2 December 2026 the AI Omnibus adds a ninth prohibited practice, AI systems that generate non-consensual sexually explicit content or child sexual abuse material. According to the Commission's AI Act Service Desk, the same date is the transitional deadline for certain providers of generative AI systems already placed on the market before 2 August 2026 to comply with the Article 50(2) marking duty.
  • In a Help Net Security interview, Veeam Field CTO Edwin Weijdema said simulated phishing and vishing exercises that use AI-generated content or cloned voices are not automatically exempt from the transparency requirements, and that the reasoning should be documented with legal and compliance input.

What it means for your organisation

Enforcement now covers general-purpose model providers and transparency duties that reach everyday deployments such as customer chatbots and generated media. Security testing is affected too: as Veeam's Edwin Weijdema pointed out, AI-generated lures and cloned voices in social engineering exercises are not automatically exempt from the transparency rules, so the legal position needs to be documented.

What we check

  • Check whether an inventory of AI systems and models in use exists, classified as GPAI-based, Article 50 transparency-relevant or potential Annex III high-risk, each with an owner.
  • Test customer-facing chatbots and AI agents for AI disclosure, prompt injection, data leakage and unsafe tool use, and keep the results as compliance evidence.
  • Review logging for LLM applications and agents so each action can be traced to an identity, input and approval ahead of the December 2027 high-risk duties.
  • Review social engineering engagement plans that use synthetic voices, deepfakes or AI-generated lures for legal sign-off, disclosure and debrief arrangements.

Sources

Services that cover these checks

  • AI & LLM application testing

    Adversarial testing of features built on large language models: chat assistants, retrieval-augmented search, agents with tool access, and generated output that reaches users or other systems. We test whether the model can be steered into leaking data, taking actions it should not, or passing untrusted output into the rest of your stack.

    Duration: 4 to 10 working days

  • Social engineering assessment

    Phishing, vishing and physical pretexting run against your staff, help desk and finance team, only under written authorisation that names the targets, pretexts and limits. Results are reported by channel and pretext, never by name: how far a convincing approach got, how many people reported it and how long the first report took to reach someone who acted.

    Duration: 2 to 4 weeks including campaign preparation

  • Threat modelling

    Facilitated workshops that map how a system or feature could be attacked before it is built or shipped. We draw the data flows with your engineers, walk them with STRIDE, rank the threats by likelihood and impact, and turn the result into backlog items and test cases your team owns.

    Duration: 2 to 5 days per system, remote or on site

Run these checks against your own systems

We pick the checks that apply to your systems. A 45-minute scoping call ends with a fixed quote and a proposed start date; the written scope follows by email.

Scope an engagementHow the scoping call works