South Africa AI Law Deadlines

The EU AI Act enters into force and becomes enforceable on August 2, 2026 — a fixed, non-negotiable deadline applicable across all 27 EU member states and the EEA. This is not a deadline for planning or preparation; it is the date on which regulators and private parties can begin filing complaints, triggering investigations, and seeking penalties. Unlike GDPR, which had a 2-year grace period before enforcement (May 2016 adoption, May 2018 enforcement), the EU AI Act's enforcement timeline is compressed: the directive was adopted in December 2023, and enforcement begins August 2, 2026 — less than 20 months from adoption to active penalty risk. Businesses should treat August 2 as the same kind of hard cutoff as GDPR's May 25, 2018: compliance is required from day one, and non-compliance creates daily exposure.

Within the August 2, 2026 deadline, nested compliance deadlines apply to different system categories. Prohibited AI systems (facial recognition in law enforcement, social credit scoring, subliminal manipulation) cannot be deployed at all after August 2 — deployment is not a compliance violation, it is an immediate prohibited activity. High-risk systems must have completed conformity assessments (including bias and fairness testing) before August 2, must document human-review processes, and must implement transparency disclosures to affected individuals by the enforcement date. Limited-risk systems must provide transparency disclosures by August 2. Prohibited systems already in deployment must be removed by August 2. The deadline structure means a single non-compliance — one high-risk system without pre-deployment assessment, one prohibited system still in operation, one missing disclosure notice — creates measurable enforcement exposure on day one of the deadline.

National implementation and enforcement begin immediately on August 2, 2026. Each EU member state has designated a national AI authority responsible for complaint intake, investigation, and penalty assessment. In South Africa, [National AI Authority details]. These authorities are staffed and equipped to begin enforcement on August 2 — they do not wait for national regulations to be finalized or enforcement guidance to be published. The first enforcement actions will likely target the highest-impact, most visible violations: prohibited systems still in operation, high-risk hiring systems without documented assessment, and chatbots that fail to disclose AI use. Smaller or less visible violations will be addressed through complaint-driven enforcement, but the baseline assumption should be that enforcement begins immediately.

The most urgent remediation timeline is now until August 2, 2026. Businesses should immediately: conduct a complete inventory of all AI systems, identify high-risk and prohibited systems, remove any prohibited systems from production, complete conformity assessments for high-risk systems, implement transparency disclosures, establish human-review processes, and document all compliance activities. For businesses with high-risk systems already in production, this timeline is extremely compressed — you must complete risk assessment, bias testing, and remediation of identified issues within months, not years. Attempting to compress months of work into weeks creates quality risk (assessment documents become rubber stamps, bias testing becomes superficial) that invites regulatory scrutiny.

After August 2, 2026, compliance obligations become ongoing and permanent. You must maintain conformity assessments for high-risk systems, re-assess annually, conduct bias re-testing after model updates, monitor human-review patterns and escalation rates, maintain audit logs, handle individual requests for explanation and appeal, and respond to regulatory inquiries. The deadline of August 2 is not the end of compliance work — it is the beginning of continuous compliance as a permanent operational function.