FOR AI AGENT BUILDERS
Your agent crosses borders. The AI Act already drew them.
In 2026, AI agents watch screens, read inboxes, chain across LLMs, and execute decisions on their own. Four things dissolve when they do — and four articles of the EU AI Act already cover each one.
We map your agent to the position the regulation has already assigned it. No interpretation. No certification. No fork.
COMPATIBLE WITH
Framework-agnostic by design. The system prompts in our public repo paste into any LLM environment that accepts a system message. None of the platforms below are affiliated with Sprinkling Act.
WHY THIS PAGE
The regulation was written before the products were shipped.
Google launched Cowork Agent: it browses the web, reads your Gmail, parses your Calendar, executes full tasks. Nik Shevchenko shipped Omi for Desktop: it watches your screen and listens to your conversations continuously. Cursor 3.1 introduced tiled multi-agent workflows: a single decision now emerges from a chain of three to five LLMs.
None of these products were built with the EU AI Act open on the desk next to them. All of them fall inside its perimeter. This page exists because nobody else is reading the regulation from the agent builder's seat.
THE 4 ACTS
Four things dissolve. Four articles cover them.
Agents don't break the law in new ways. They concentrate four old failure modes into a single product. Each ACT names one. Each maps to specific articles of the EU AI Act. The acronym mirrors the regulation itself: the 4 ACTS of the AI Act.
Accountability
When the agent acts, who answers?
Dissolves when: The agent executes decisions on a human's behalf without explicit per-action authorization.
Articles: Art. 14 (human oversight) · Art. 26 (deployer obligations) · Art. 6 + Annex III
Gate: Routes to Gate 03 — High-Risk Sector
Example: Google Cowork Agent books your trip. If it books wrong, who is the deployer under Art. 26 — you, your employer, or Google?
Consent
Who agreed to be captured?
Dissolves when: The agent runs always-on capture (screen, audio, inbox) without per-session consent surface.
Articles: Art. 5 (manipulation, prohibited practices) · Art. 50 (transparency) · GDPR intersection
Gate: Routes to Gate 01 if Art. 5 · Gate 05 if Art. 50
Example: Omi for Desktop watches every screen and listens to every conversation. The third party in the room never opted in. Art. 50 transparency obligations are triggered on the first sentence captured.
Traceability
What can be reconstructed after the fact?
Dissolves when: A decision emerges from a chain of multiple LLMs and tool calls without a single auditable log surface.
Articles: Art. 12 (record-keeping) · Art. 13 (transparency to deployers) · Art. 15 (accuracy and robustness)
Gate: Routes to Gate 03 when chain affects Annex III · Gate 02 if safety component
Example: Cursor 3.1 tiled multi-agent workflows: a single code decision emerges from five LLMs talking to each other. Which one's log do you produce when an auditor asks why the agent merged that pull request?
Skill
Does the person shipping understand what they shipped?
Dissolves when: The builder lacks the technical or regulatory literacy to evaluate the system they put in production.
Articles: Art. 4 (AI literacy) · Art. 26 (deployer due diligence obligations)
Gate: Cross-cutting — affects deployer obligations on every gate
Example: A founder ships a vibe-coded agent built in 6 months from tutorials. The agent processes EU customer data. Art. 4 requires documented AI literacy. There is none.
These four ACTS are not a new methodology. They are a translator. Each one points to gates already documented in our public methodology. If your agent triggers any of them, the regulation already has an answer for you. The free diagnostic surfaces which.
THE BRIDGE
How the 4 ACTS connect to our 6 Gates.
The ACTS framework is not a replacement for our scoring methodology. It is a translator. Every ACT routes to one or more of the 6 regulatory gates documented in our public methodology. If your agent triggers an ACT, the gate it lands on determines the position your Full Report will assign.
A · Accountability
Art. 14, Art. 26, Art. 6 + Annex III
Gate 03 — High-Risk Sector
Most autonomous-action agents land in Annex III.
C · Consent
Art. 5, Art. 50, GDPR intersection
Gate 01 if Art. 5 · Gate 05 if Art. 50
Ambient capture either ends the assessment or triggers transparency obligations.
T · Traceability
Art. 12, Art. 13, Art. 15
Gate 03 when chain affects Annex III · Gate 02 if safety component
Chained agents inherit the gate of the riskiest decision in the chain.
S · Skill
Art. 4, Art. 26
Cross-cutting — affects deployer obligations on every gate
Builder literacy is not a gate, it is a precondition for every gate.
WHAT THE FIELD LOOKS LIKE
Five patterns we currently see in production agents.
These five patterns are not theoretical. Each one is visible in shipping products as of April 2026. Each one routes to one or more ACTS, and from there to a specific gate position.
Ambient capture
ACT COmi for Desktop, NotebookLM
Always-on screen and audio capture without per-session consent.
Autonomous action
ACT AGoogle Cowork Agent, Lovable Payments
The agent decides and executes without explicit per-action authorization.
Multi-agent chaining
ACT TCursor 3.1 Tiled, Anthropic Routines
A single output emerges from a chain of 3-5 LLMs talking to each other.
Vibe-coded shipping
ACT SBuilders from 6-month tutorials
The builder ships an agent they cannot evaluate technically or legally.
Second-brain sync
ACT C + Tmem_base, NotebookLM + Obsidian
Personal knowledge auto-synced with LLM providers — the sovereignty boundary dissolves.
WHERE YOU STAND
Don't guess. Run the 9-question diagnostic.
The free diagnostic applies the 6 gates to your system in 60 seconds. If you build an AI agent — autonomous action, ambient capture, multi-agent chain, or any combination — the diagnostic returns the position the regulation has already assigned it. No account, no email, no funnel before the answer.
WHAT THE FULL REPORT ADDS
ACTS is already inside every Full Report.
Every Full Report (€690) is already an ACTS-aware analysis. The 4 ACTS are not a separate add-on — they are the lens we use to read your system before assigning a gate. When the report determines that your agent triggers Gate 03 because it executes autonomous decisions on behalf of a user, that is ACT A (Accountability) being applied. When it flags an Art. 50 transparency obligation because of always-on capture, that is ACT C (Consent). The report names the gate. ACTS names why.
What the Full Report containsAI POSITIVE FOR AGENTS
ACTS surfaces exposure. AI Positive measures performance.
ACTS surfaces your regulatory exposure. AI Positive measures your ethical performance. The two are complementary, not redundant. An agent can be ACTS-clean (no gate triggered, low risk position) and still score poorly on AI Positive because it displaces more human judgment than it augments. Conversely, a high-risk Annex III system can score high on AI Positive if it is built with strong governance, explainability, and reversibility. ACTS asks: are you legal? AI Positive asks: are you good? Both questions matter. Both answers come from the same Full Report.
Read about AI PositiveDISPLAY THE BADGE
If you've integrated the prompts, show it.
If you've integrated one or more of the ACTS system prompts into your AI agent, you may display this badge in your repository's README. It is a self-declaration of awareness — not a certification, not a verification, not an audit. The badge wording matters: read the disclaimer in the GitHub repo before you use it.
Copy this Markdown into your README:
[](https://sprinklingact.com/ai-agents)WHAT ACTS IS NOT
ACTS is not a certification. It is not a standard you can apply for. It is not an open-source framework you can fork on GitHub. It is the lens we use to read the EU AI Act for teams that ship AI agents in 2026. The deliverable is the Full Report. The verifiable artefact is the date on the document. We are not here to license a brand. We are here to assign positions to systems that don't yet know what position they hold.
SEE ALSO
The 6 Gates Methodology
How we score AI Act position. Article by article.
AI Positive
Compliance is the floor. AI Positive is the ceiling.
Free Diagnostic
9 questions, 60 seconds, instant result.
Full Report
What the €690 report contains. 15-22 pages.
Verify a badge
Public registry of SA-XXXX assessed agents. Look up an existing badge or get yours.