Scorecard/Goodfire

Goodfire

Data gathering in process

AI interpretability research lab using mechanistic interpretability to understand, audit, and steer AI model behavior for safer deployment.

HQUS
Est2024
Size11-50
EU AI ActLimited Risk
goodfire.ai
Score
56.3 / 100
Evidence
6 items

Strong safety posture with established governance frameworks and active risk management.

Strengths:Governance Maturity, Technical Safety, Risk Assessment, External Engagement
Weaknesses:Regulatory Readiness
Focus Areas
interpretabilityai safetymodel auditingalignment research

Security Assessment

Security-relevant indicators for vendor evaluation

Security Posture
64
TS-01dim: 72
Red Teaming & Pre-deployment Testing
Adversarial testing before deployment
TS-05dim: 72
Robustness & Adversarial Resilience
Resistance to adversarial attacks
RA-01dim: 55
Sector-Specific Risk Assessment
Risk analysis for deployment context
RA-03dim: 55
Dual-Use & Misuse Risk
Dangerous capability awareness
RA-07dim: 55
Incident History & Track Record
Past incidents and response quality
EE-04dim: 75
Vulnerability Disclosure Program
Bug bounty or CVE reporting process
Incident History
Goodfire incident records sourced from AIAAIC Repository and public reporting.
Integration: AIAAIC, OECD AI Incidents Monitor
Third-Party Audits
External audit reports, SOC 2 attestations, and ISO certifications verified where published.
Sources: Company filings, registry lookups
CVE & Disclosures
Known vulnerabilities and security advisories from NVD, GitHub Security Advisories, and vendor pages.
Sources: NVD, GHSA, vendor disclosure pages

Dimension Breakdown

GM
Governance Maturitymedium
Published policies, corporate structure, safety mandate, whistleblowing, executive commitment.
55
1 evidence items
GM-01
TS
Technical Safetymedium
Benchmarks, adversarial robustness, fine-tuning safety, watermarking, model cards, research output.
72
2 evidence items
TS-01TS-02
RA
Risk Assessmentlow
Dangerous capability evaluations, thresholds, external testing, bug bounty, halt conditions.
55
1 evidence items
RA-01
RR
Regulatory Readinesslow
ISO 42001, EU AI Act compliance, GPAI obligations, international commitments, incident reporting.
30
EE
External Engagementmedium
Survey participation, research support, transparency, behavior specs, open-source contributions.
75
2 evidence items
EE-01EE-02

Social Impact & Safety Profile

Moderate

Goodfire builds interpretability tools that help researchers and developers understand what AI models are doing internally. This directly supports fairness auditing, bias detection, and accountability. While the social impact is implicit in the product's purpose, formal social impact policies are not yet published.

interpretabilityfairness auditingaccountability
Why it matters for safety

If you cannot understand what a model is doing internally, you cannot verify it is safe. Interpretability is the foundation of safety verification - without it, safety claims are based on behavioural testing alone, which can be gamed.

Civilizational Risk Awareness

2/3

Genuine safety motivation evident in team background and mission. Governance structure does not yet encode catastrophic risk priority over commercial pressure.

Responsible Scaling Policy

None

No published RSP or equivalent. As a tooling company (not a model developer), an RSP in the traditional sense may not apply. However, no equivalent framework exists for how interpretability tooling capabilities should be gated or governed as they scale.

RSPs are designed for frontier model developers. For safety tooling companies, the equivalent question is: does the company have a policy governing how its tools are used and by whom? Does it restrict access to interpretability insights that could be used to circumvent safety measures?

Mission Drift Protection

1/3
  • Mission statement in company charter
  • No PBC status
  • No capped profit structure
  • No independent safety board
  • No charter provision preventing mission change without supermajority

Vulnerability Disclosure

None

No CVD programme for AI safety vulnerabilities. Relevant vulnerabilities would include: misuse of interpretability insights to bypass model safety measures, or failures in tools that give false confidence in model safety.

Most safety startups have no CVD - this is an ecosystem-wide gap.

Safety Reporting

- None
Research blog postsirregular

No structured safety reporting. Blog posts provide insight into research progress but do not constitute systematic safety transparency. For a company building safety tooling, structured reporting on tooling performance would significantly strengthen credibility.

Dual-Use Risk

LowAI×Cyber Offensive

Low but non-zero dual-use risk. The company should have a documented position on access controls for interpretability insights.

Mitigation details
Interpretability insights require significant technical expertise to exploit
Access controls on tooling unclear
No published dual-use assessment
No documented position on who should access interpretability insights

Recent Signals

View all signals

Grants, funding rounds, policy updates, and market events linked to Goodfire.

Funding
15 Jan 2025
Goodfire raises $50M Series A for interpretability tools

Enterprise interpretability platform backed by Lightspeed Venture Partners.

interpretability$50M

Need a detailed report for Goodfire?

Subscribe to express interest in indicator-level evidence, peer benchmarking, and regulatory gap analysis - or reach out to request a full company overview brief.