AI Pulse Daily Brief | 2026-07-29
Reading time ~2 mins
- The EU's amendment package to the AI Act entered into force on 27 July, moving the high-risk deadlines to December 2027 and August 2028 without moving the classification work behind them.
- Amazon's managed service for running AI agents now isolates each agent's logs and encryption by default, but only for agents created after 20 July.
Top signal
The EU's AI Act amendment is in force, moving the high-risk deadlines to December 2027 and August 2028. Authority
The European Commission announced on 27 July that the AI Omnibus, its package of amendments to the AI Act, has entered into force across the EU. Stand-alone high-risk AI systems, the ones listed in the Act's Annex III, now come under the rules from 2 December 2027. High-risk AI built into regulated physical products, covered by Annex I, follows from 2 August 2028. The same notice extends simplified obligations to small mid-cap companies, widens access to regulatory sandboxes, and changes the central-database and post-market-monitoring arrangements.
The dates moved; the work behind them did not. Deciding which of the two deadlines applies to a given system still rests on a classification judgement that the Omnibus leaves untouched. The sandbox, database and monitoring changes also take effect ahead of both deferred dates, so a roadmap built to the previous timetable is now out of order rather than simply early. This is the second AI Act timing change to land in a week, and it is a separate one: the transparency duties that start on 2 August 2026 are not addressed here.
Innovation
Amazon gives each AI agent its own isolated log by default, but only for agents built after 20 July. Vendor
Amazon Web Services said on 23 July that its managed service for running AI agents, Bedrock AgentCore, now gives each agent its own separate log. That log holds the agent's prompts, traces and output, and carries its own access permissions and encryption under a key the customer controls. The arrangement is on by default only for agents created from 20 July onwards. Agents built before that date keep the previous shared setup until someone changes a runtime setting and upgrades a monitoring component. The audit and encryption gap therefore sits with the earliest pilots rather than the newest work, closing quietly for new agents while staying open on the deployments already furthest along.