Compliance frameworks

Every framework your auditor
cares about. Pre-built.

EU AI Act, NIST AI RMF, ISO 42001, OWASP LLM Top 10, SOC 2, HIPAA, PCI DSS. Install a pack. Guard enforces it. Audit log is automatic.

Conduct makes compliance structural โ€” not documented after the fact, enforced before execution. Every allowed action produces a Governance Authorization Artifact: replay-verifiable evidence that authorisation was confirmed before execution. Custody proof, not a reconstructed log.

๐Ÿ‡ช๐Ÿ‡บ

EU AI Act

Regulation10 rules
Type: RegulationยทMandatory: MandatoryยทBuyer: Any EU-market AI system

The EU AI Act bans social scoring, emotion recognition in workplaces, mass surveillance, and subliminal manipulation. ConductGuard enforces these prohibitions at the point of code โ€” before the agent runs.

Install pack โ†’

Control coverage

Article 5.1aSubliminal manipulation โ€” blocked
Article 5.1bBiometric categorization โ€” blocked
Article 5.1cSocial scoring โ€” blocked
Article 5.1fEmotion recognition in workplaces โ€” blocked
Article 5.1hMass surveillance โ€” blocked
Article 10Data governance โ€” PII in training blocked
Article 12Logging โ€” audit log enforced
Article 13Transparency โ€” disclosure required
Article 14Human oversight โ€” override blocked
Article 53GPAI high-risk use โ€” warned
๐Ÿ›๏ธ

NIST AI RMF

Risk Framework10 rules
Type: Risk frameworkยทMandatory: Voluntary (de facto)ยทBuyer: Federal agencies

The NIST AI Risk Management Framework structures AI risk across four functions. ConductGuard ships one rule per key control โ€” blocking policy bypass, flagging high-risk deployments, and alerting on disabled monitoring.

Install pack โ†’

Control coverage

GOVERN 1.2Least privilege โ€” escalation blocked
GOVERN 2.2Policy bypass โ€” blocked
GOVERN 4.1Accountability logging โ€” enforced
GOVERN 6.1Human oversight โ€” override blocked
MAP 1.5Unverified data sources โ€” warned
MAP 5.1High-risk domain deployment โ€” warned
MEASURE 1.1Monitoring disabled โ€” blocked
MEASURE 2.5Error swallowing โ€” flagged
MANAGE 1.3Third-party AI vendors โ€” audit required
MANAGE 4.1Data retention โ€” policy enforced
๐Ÿ“

ISO 42001

Standard10 rules
Type: AI management systemยทMandatory: VoluntaryยทBuyer: Boards, compliance teams

ISO 42001 is the international management system standard for AI. ConductGuard operationalises Clauses 8โ€“10: access control, bias testing, data quality, impact assessment, responsible use, and incident capture.

Install pack โ†’

Control coverage

Clause 8.1Operational controls and human oversight
Clause 8.1Access control and supplier relationships
Clause 8.2Bias testing โ€” unvalidated models flagged
Clause 8.4Data quality โ€” unlabeled data warned
Clause 8.4Data minimisation โ€” PII in training blocked
Clause 8.5Impact assessment โ€” high-risk use flagged
Clause 8.6Responsible use โ€” off-label use blocked
Clause 8.6Prohibited practices alignment
Clause 9.1Monitoring โ€” disable attempt blocked
Clause 10.2Incident capture โ€” error swallowing blocked
๐Ÿ›๏ธ

IRS Publication 1075

Government8 rules
Type: Federal regulationยทMandatory: MandatoryยทBuyer: Govt agencies + contractors handling tax data

IRS Publication 1075 governs Federal Tax Information (FTI) under IRC Section 6103. Any agency or contractor handling IRS-shared tax data must safeguard it from disclosure, improper use, and inadequate storage. ConductGuard blocks EINs, AGI values, and W-2/1099 data from entering AI prompts, training datasets, or external endpoints.

Install pack โ†’

Control coverage

Sec 4 / IRC 6103EIN in AI prompt โ€” blocked
Sec 4 / IRC 6103Tax return data in model context โ€” blocked
Sec 4FTI to external AI endpoint โ€” blocked
Sec 4FTI in training datasets โ€” blocked
Sec 9FTI in unencrypted files โ€” blocked
Sec 4 / Sec 10FTI unredacted in logs โ€” blocked
Sec 4Third-party AI FTI processor โ€” warned
IRC 6103 / Sec 4FTI beyond authorised purpose โ€” blocked
๐Ÿ”

OWASP Top 10 for LLMs

Security20 rules
Type: Security standardยทMandatory: VoluntaryยทBuyer: Security teams, enterprise customers

OWASP's LLM Top 10 covers prompt injection, insecure output handling, training data poisoning, and supply chain vulnerabilities. ConductGuard maps rules to each identifier so your security team speaks the same language as your auditor.

Install pack โ†’

Control coverage

LLM01Prompt injection โ€” blocked
LLM02Insecure output handling โ€” flagged
LLM03Training data poisoning โ€” warned
LLM04Model DoS โ€” rate patterns blocked
LLM05Supply chain vulnerabilities โ€” warned
LLM06Sensitive information disclosure โ€” blocked
LLM07Insecure plugin design โ€” flagged
LLM08Excessive agency โ€” override blocked
LLM09Overreliance โ€” human oversight enforced
LLM10Model theft โ€” exfiltration blocked
๐Ÿ“‹

SOC 2

Audit4 rules
Type: Security auditยทMandatory: VoluntaryยทBuyer: Enterprise customers

SOC 2 auditors look for evidence of control. ConductGuard produces that evidence automatically โ€” blocking hardcoded secrets, warning on PII in logs, and flagging debug mode in production.

Install pack โ†’

Control coverage

CC6.1Logical access โ€” hardcoded secrets blocked
CC6.3Least privilege โ€” escalation blocked
CC7.2System monitoring โ€” PII in logs warned
CC8.1Change management โ€” debug mode in prod flagged
๐Ÿฅ

HIPAA

Healthcare3 rules
Type: Federal regulationยทMandatory: MandatoryยทBuyer: Healthcare + business associates

HIPAA ยง164.312 requires technical safeguards for electronic PHI. ConductGuard blocks patient IDs, SSNs, and DOBs from entering source files and flags unencrypted transmission patterns.

Install pack โ†’

Control coverage

ยง164.312(a)Access control โ€” PHI access blocked
ยง164.312(b)Audit controls โ€” PHI in source flagged
ยง164.312(e)Transmission security โ€” unencrypted PHI warned
๐Ÿ’ณ

PCI DSS

Finance3 rules
Type: Industry standardยทMandatory: Mandatory (card brands)ยทBuyer: Any entity processing payments

PCI DSS Requirements 3 and 4 prohibit storing card data in source and mandate strong encryption in transit. ConductGuard enforces both at the code layer โ€” no post-commit remediation needed.

Install pack โ†’

Control coverage

Req 3.4PAN in source โ€” blocked
Req 3.5CVV in source โ€” blocked
Req 4.2Weak TLS / cleartext โ€” flagged

Where regulators have not caught up yet

Interim controls, aligned to the underlying principles.

Where regulators have not yet published agent-specific guidance, Guard provides interim controls aligned to the underlying principles. Ready when the eventual guidance arrives.

US banking

SR 26-2 (April 2026) explicitly scoped agentic AI out as novel and rapidly evolving. Agent-specific guidance is signaled but not yet issued. 42 percent of financial firms are already using or assessing agents. Guard aligns to SR 11-7 pillars via the conduct-financial-services pack.

See the financial services solution โ†’

Life sciences

FDA has not yet issued agent-specific guidance. Validation teams reconcile FDA CSA, FDA/EMA GMLP, ICH Q9, GAMP 5, and ISO 42001. Guard applies the underlying principles at the runtime layer via the conduct-life-sciences pack.

See the life sciences solution โ†’

EU AI Act (Article 26)

High-risk obligations are phasing in through 2026. Article 26 requires human oversight logs. Guard produces the decision-log evidence continuously and automatically, mapped to Article 26 across every applicable pack.

See the EU AI Act pack โ†’

Your auditor wants evidence.
Guard produces it automatically.

Install a compliance pack. Every agent decision is logged as a Governance Authorization Artifact, mapped to a framework control, and exportable. Structural compliance โ€” not documented after the fact.

Free tier ยท No infrastructure changes ยท Works in minutes

ConductAI: Runtime Governance for AI Agents