AI in May 2026: Europe eases the AI Act timeline as agents get to work
May brought welcome clarity on European regulation, alongside a steady stream of new models and agent tools from the largest AI providers. The common thread was practical: more time to prepare for compliance, and more ways to put agents to work safely. Here are the developments worth knowing about, and what they mean for your organisation.
The EU agrees to delay key AI Act obligations
On 7 May, the Council of the EU and the European Parliament reached a provisional agreement on the "Digital Omnibus on AI", a package designed to simplify and streamline the AI Act. The biggest change is timing. Rules for stand-alone high-risk AI systems, such as those used in recruitment, credit scoring and biometric identification, will now apply from 2 December 2027 rather than August 2026. Rules for AI built into regulated products move to 2 August 2028.
The deal also adds a new prohibition on AI systems that generate child sexual abuse material or non-consensual intimate images, and delays watermarking requirements for AI-generated content to December 2026. Some relief previously reserved for small businesses is extended to small mid-cap companies, and the duty on organisations to ensure staff AI literacy is softened into an obligation on the Commission and member states to support it. The agreement still needed formal approval from both institutions.
What it means for your business: If you use AI in hiring, lending or other high-risk areas and trade with the EU, you have more time, but not a reprieve. The obligations themselves have not gone away. Use the extra months to document how your systems work, what data they use and who oversees them, so compliance becomes routine rather than a last-minute scramble. Training staff to use AI well remains good practice, whatever the legal wording.
Google unveils Gemini 3.5 and a wave of agents at I/O
At its I/O developer conference on 19 and 20 May, Google launched Gemini 3.5, which it describes as combining "frontier intelligence with action", with new capabilities for complex, multi-step agentic workflows. Gemini 3.5 Flash became the default model in the Gemini app and Google Search, with the more powerful Gemini 3.5 Pro due to follow. Google also introduced Gemini Omni, a model that can create from any input, starting with video.
Alongside the models came several agents, including Gemini Spark, a personal agent that keeps working in the background, a Daily Brief that summarises Gmail, Calendar and Tasks each morning, and information agents in Search that monitor topics on a user's behalf. Google also said voice-driven Docs Live and Gmail Live would arrive over the summer.
What it means for your business: Many of these features are aimed at consumers first, but they shape what your staff and customers will expect from everyday software. Background agents that monitor information and prepare briefings are a pattern worth borrowing internally, for example for tracking competitor news, supplier updates or key account activity.
UiPath brings coding agents into governed automation
On 12 May, UiPath announced UiPath for Coding Agents, which it describes as the first business orchestration and automation platform with native integration for coding agents. It initially supports Anthropic's Claude Code and OpenAI's Codex, allowing automations generated by AI to run on UiPath's platform with the same policy enforcement, audit trails, credential management and role-based access as traditionally built ones. UiPath says business analysts and process owners can now prototype automations through natural-language conversations.
Chief executive Daniel Dines said the platform "treats AI-generated automations as first-class citizens".
What it means for your business: AI can now write automation scripts in minutes, but unmanaged scripts quickly become a risk. This move points to a sensible middle ground: let people closer to the process build automations quickly, while keeping them inside a governed platform with proper logging and access controls. If you already use RPA, review how AI-generated automations will be approved, monitored and maintained.
Anthropic releases Claude Opus 4.8
On 28 May, Anthropic released Claude Opus 4.8, just over a month after its predecessor. According to TechCrunch, the headline improvement is that the model is more likely to flag uncertainties about its work and less likely to make unsupported claims. Anthropic also launched a research preview of "dynamic workflows", which lets Opus coordinate hundreds of parallel sub-agents on large tasks such as code migrations. Pricing was unchanged. Anthropic added that it expected to bring "Mythos-class" models, its more capable tier first previewed in April, to all customers in the coming weeks.
What it means for your business: A model that says when it is unsure is often more useful than a slightly cleverer one that sounds confident when it is wrong, particularly in finance, operations and customer communications. When testing models, measure how often they flag uncertainty and how accurate they are, not just how impressive the output looks.
Practical takeaways
If you use AI in hiring, lending or other high-risk areas, start documenting those systems now, using the extended EU deadlines as a planning window.
Keep investing in staff AI training, even though the AI Act's literacy duty has been softened.
Pilot a simple monitoring or daily-briefing agent for a team that spends time gathering information.
Set clear rules for how AI-generated automations are reviewed, approved and maintained before they go live.
Include accuracy and honest handling of uncertainty in any model evaluation, not just speed and output quality.
If you would like help making sense of the new EU timeline, or working out where agents and intelligent automation could save your teams time, we would be happy to talk it through. Book a call with the Manta Global AI team and we can explore what is possible for your organisation.
Comments