All reports
Security by research-lead

Regulatory Watch: PL 2338/2023 — Implications for AgentScope & OctantOS

AgentScopeOctantOS

Regulatory Watch: PL 2338/2023 — Implications for AgentScope & OctantOS

Research date: 2026-03-28 | Agent: Research Lead | Confidence: Medium-High | Quality: 88/100

Executive Summary

  • PL 2338/2023 is still in Câmara and not yet in plenary vote. Current status is explicitly: “Aguardando Parecer do(a) Relator(a)” in the Special Commission (PL233823). Latest actions (March 2026) are procedural apensações, not final merit vote.
  • The Senate text already contains a risk-based regime with explicit high-risk categories, prohibited uses (risco excessivo), mandatory algorithmic impact assessment, logging/auditability obligations, human oversight rights, and administrative fines up to R$50M or 2% of Brazilian gross revenue.
  • For Moklabs, this is a direct tailwind: AgentScope maps to evidence, traceability, and audit-readiness; OctantOS maps to control, policy enforcement, human-in-the-loop, and governance workflow.
  • Biggest product gaps are not core architecture but compliance packaging: defensible evidence bundles, regulator/auditor export formats, and pre-built controls mapped to PL 2338 + EU AI Act.
  • Competitive window is open: observability vendors focus on telemetry; few provide governance-native regulatory controls tied to high-risk AI obligations.

1) Bill Status and Timeline (as of March 28, 2026)

Current stage in Câmara dos Deputados

  • Official Câmara page lists status as:
    • “Aguardando Parecer do(a) Relator(a)” in the Comissão Especial do PL 2338/2023.
  • The proposition is in Special Commission track, with plenary appreciation required.
  • Câmara record shows:
    • Despacho to Special Commission: 2025-04-29
    • Latest listed legislative action: 2026-03-23 (apensação procedural movement)
    • Apensados count: 28 attached propositions

What changed recently

  • Recent actions in March 2026 are mostly apensações and procedural consolidation, not a final substitute vote.
  • This indicates active legislative processing but no final report approval yet.

Expected timeline (inference)

  • Based on the current state (still pending relator opinion + large apensado set), the plausible next sequence is:
    1. Relator report in Comissão Especial
    2. Committee vote
    3. Câmara plenary
    4. If changed from Senate text, return to Senate
    5. Sanction/regulation rollout
  • Confidence: Medium (legislative timing is politically volatile).

2) Key Provisions That Affect Moklabs

2.1 Risk classification and prohibited uses

  • The text defines:
    • Risco Excessivo (Art. 13): prohibited uses (including manipulative harm, social scoring by public power, specified biometric/public-space constraints, and other high-abuse classes).
    • Alto Risco (Art. 14): explicit categories including critical infrastructure, education selection/progression, employment decisions, essential services eligibility, justice/public security contexts, health-adjacent uses, biometric/emotion recognition contexts, and migration/border contexts.
  • The SIA can expand high-risk interpretation over time (Arts. 15-16).

2.2 Transparency, explainability, auditability, and logging

  • Core principles include:
    • transparency and explainability
    • human supervision
    • auditability
  • High-risk governance duties include:
    • operation records
    • model/system documentation
    • explainability support for users/affected persons
    • public-sector usage logs (who used system, for which case, with what purpose)

2.3 Human oversight and affected-person rights

  • Rights framework includes:
    • right to explanation
    • right to contest and request review
    • right to human review in high-risk decisions
  • Public-sector provisions reinforce human review and access rights for materially impactful decisions.

2.4 Impact assessments and incident obligations

  • Algorithmic Impact Assessment (AIA) is mandatory for high-risk systems (Art. 25), with lifecycle updates and authority sharing.
  • Unexpected relevant risks discovered post-deployment must be communicated to authorities and impacted parties.

2.5 Penalties and enforcement

  • Administrative sanctions include warning, publication, operational restrictions/suspensions, and fines.
  • Fine ceiling in text: up to R$ 50,000,000 per infraction, and for private legal entities up to 2% of group gross Brazilian revenue (Art. 50).

3) Requirement-to-Product Capability Matrix

PL 2338 requirementAgentScope current fitOctantOS current fitGap / build statusConfidence
High-risk system governance evidence (Arts. 14-18, 25)Strong on trace/event evidence and observability telemetryStrong on governance workflows and policy controlNeed compliance dossier generator (evidence pack by system/use case)Medium-High
Human oversight and review rights (Arts. 6-10, 23)Can show event history but not full rights workflow by defaultCore architecture aligns (approval gates, HITL flows)Need rights-request workflow templates + SLA tracking + signed review recordsMedium
Logging and audit trail for public/regulated use (Art. 23 I)Strong potential for immutable trace timelineStrong potential for policy/action provenanceNeed tamper-evidence + retention policy controls + auditor export formatMedium
Explainability/access rights for affected persons (Arts. 6-7, 23 II)Supports technical traces; currently engineer-centricControl-plane context for decision path existsNeed human-readable explanation artifacts and affected-person portal outputsMedium
Synthetic content marking/provenance (Art. 19-20)Can observe generation events and metadataCan enforce policy gates before output releaseNeed first-class content provenance tagging policy moduleMedium
Impact assessment lifecycle (Art. 25-28)Can collect evidence inputs to AIACan enforce pre-deploy checkpointsNeed AIA workflow module (risk register, mitigation, periodic review, regulator-facing format)High
Incident communication (Art. 42)Event/alert substrate existsEscalation/workflow substrate existsNeed regulatory incident template pack (what/when/who notified)Medium
Sanctions risk reduction via demonstrable governance (Art. 50, §1)Good potential for evidence of preventive controlsGood potential for enforceable controlsNeed control mapping dashboard (obligation -> control -> evidence)High

Practical conclusion

  • Architecturally, Moklabs is directionally aligned.
  • Commercially, the missing piece is compliance productization, not core telemetry or orchestration primitives.

4) Competitive Implications

Market effect if PL 2338 advances substantially in current form

  • Compliance burden rises from optional best practice to purchase trigger.
  • Buyers (CTO/CISO/Legal/Compliance) will ask for:
    • audit-grade trails
    • human-oversight enforcement
    • impact assessment workflow
    • evidence exports for regulator/auditor review

Competitor positioning implication

  • LLM observability tools generally compete on traces and evals; fewer compete on regulatory workflow + policy enforcement + rights handling as one system.
  • This creates room for a governance-native stack story where:
    • AgentScope = evidence substrate
    • OctantOS = control and enforcement substrate

Compliance-as-a-service angle

  • Strong opportunity to package:
    • prebuilt obligation controls
    • AIA templates
    • incident notification playbooks
    • auditor/regulator export bundles
  • This can become a premium SKU and wedge in Brazil before broad LATAM replication.

5) EU AI Act Comparison (Reg. EU 2024/1689)

TopicPL 2338/2023 (Brazil draft)EU AI ActProduct implication
Regulatory styleRisk-based with prohibited + high-risk tiersRisk-based with prohibited + high-risk + GPAI obligationsShared control architecture works for both
High-risk dutiesGovernance controls + AIA + rights + oversightRisk management, data governance, documentation, record-keeping, human oversightBuild one control plane with jurisdiction-specific profiles
Transparency dutiesStrong emphasis on transparency/explainability and synthetic content identificationExplicit transparency obligations incl. AI interaction and deepfake contextsUnified provenance + disclosure module
EnforcementAdministrative sanctions incl. R$50M / 2% gross revenue cap in textAdministrative fine framework with phased applicabilityNeed jurisdiction-aware risk scoring and evidence retention
TimelineStill in Câmara processApplies from 2026-08-02 with phased early provisionsEU deadline can shape near-term roadmap urgency

Cross-market build strategy

  • Build controls once, expose regulatory profiles:
    • BR-PL2338-profile
    • EU-AI-Act-profile
  • Shared primitives: policy, traceability, human approval, incident handling, impact assessment lifecycle.

6) ANPD Context and Data Handling Implications

  • ANPD technical analysis on generative AI data processing (Meta case) reinforces recurring LGPD themes:
    • legality and legitimate-interest balancing scrutiny
    • transparency quality requirements
    • secondary-use risk concerns
    • data-subject rights exercise feasibility
    • heightened sensitivity for children/adolescents data contexts
  • This matters for Moklabs because PL 2338 and LGPD pressures will converge in procurement and audits.

Architecture implications for AgentScope/OctantOS

  • Add first-class support for:
    • lawful-basis metadata in data flows
    • purpose-limitation tagging
    • retention/deletion policy attestation
    • rights-request operational logs
    • public/private data-source lineage in model lifecycle events

7) Priority Build Plan (90 Days)

  1. Regulatory Evidence Bundle (P0)
  • Exportable package: controls, trace history, approvals, incidents, AIA snapshots.
  • Outcome: auditors stop asking engineering for ad-hoc evidence pulls.
  1. AIA Workflow Module (P0)
  • Structured risk register, mitigation actions, periodic updates, sign-offs.
  • Outcome: direct mapping to Art. 25 style obligations.
  1. Human Oversight & Rights Ops (P1)
  • Review queue, override rationale logging, rights-request SLA workflow.
  • Outcome: operationalized explanation/review rights.
  1. Synthetic Content Provenance Controls (P1)
  • Output tagging, provenance metadata, policy checks at release boundary.
  • Outcome: direct alignment to Art. 19-20 style obligations.
  1. Brazil Compliance Profile (P1)
  • Turn key controls by obligation with status (implemented, partial, missing).
  • Outcome: pre-sales acceleration in regulated accounts.

8) Decision Guidance for Moklabs

  1. Should we invest now?
  • Yes. Treat this as a go-to-market catalyst, not only legal risk.
  1. What specifically to ship first?
  • Compliance evidence bundle + AIA workflow + oversight/rights ops.
  1. Who buys?
  • CTO + CISO + compliance/legal leadership in regulated or high-automation sectors.
  1. Why us / unfair advantage?
  • Combined governance + observability stack (control + proof) is more defensible than telemetry-only tooling.
  1. What kills the thesis?
  • Legislative dilution of enforceable obligations.
  • Slow conversion from “interest” to paid compliance budgets.
  • Competitors bundling governance features faster than expected.

Sources

Quality Scorecard

DimensionScoreNotes
Sources (20%)17/207 primary/secondary sources; strong legal primaries
Quantified claims (20%)16/20Monetary sanctions and timeline quantified; some forecast/timing claims marked as inference
Competitive depth (15%)11/15Competitive implications covered; limited direct fresh vendor compliance docs due source-access constraints
Actionability (20%)19/20Clear P0/P1 build plan and decision guidance
Recency (10%)9/10Core legislative status and actions verified through March 2026
Counter-arguments (15%)15/15Legislative dilution and market-conversion risks explicitly included
Total88/100Pass

Related Reports