See how Penthropic verifies your vendors?
Book a Discovery CallThird-party risk management
Most third-party risk tools record the answer and file the document. Penthropic reads the SOC 2 report, the ISO certificate, the policies and the DPA the vendor uploaded, then checks every high-stakes answer against them. When the answer and the evidence disagree, you hear about it before signature, not six months after it.
Built for
The moment that matters
A vendor answers the questionnaire the way they want to be read. Their own SOC 2 report says something else. The AI reads both and puts the two sentences side by side, before you sign.
Yes
Enforced on all administrative accounts, today.
“Multi-factor authentication is enforced on all administrative accounts.”
Source: the vendor's own words, kept on the record
Q3 2026
When their own auditor expects full enforcement.
“Multi-factor authentication for privileged accounts was deployed mid-period; the auditor noted full enforcement is expected Q3 2026.”
Source: SOC 2 Type II report, uploaded by the vendor
Contradiction
How it works
Three jobs on the same vendor record. No reading PDF evidence by hand, no “yes” nobody checked, no drift you find out about at renewal.
The AI reads every SOC 2 report, ISO certificate, policy and DPA the vendor uploaded, then compares every high-weight control claim in the questionnaire with what the evidence says.
The questionnaire re-runs on a cadence, and a finding is opened the moment a high-weight control regresses. A Tier 1 vendor that turned multi-factor authentication off last quarter shows up the same day.
The attack-chain composer takes the vendor's open findings and renders a five-step, MITRE-aligned breach path. The narrative a board asks for, rather than another list of CVEs.
Every submission gets a multi-perspective review, and the four views are synthesised into one. A privacy problem and a procurement problem do not look alike, and neither is found by reading the questionnaire once.
Versus the incumbents
Most third-party risk platforms take weeks to stand up, need a professional services engagement to configure, and still leave your reviewers reading PDF evidence by hand.
| Capability | Typical TPRM | Penthropic TPRM |
|---|---|---|
| Time to first vendor live | Weeks of professional services | Same day. We send your access credential by hand, then you self-serve |
| Day-to-day interface | Heavy, multi-portal, dated | One fast interface, keyboard first |
| Editing questions, templates and rules | A consultant ticket, days to land | Edited in the app, live immediately |
| Send a questionnaire, receive answers | Yes | Yes |
| Evidence vault with expiry tracking | Yes | Yes |
| Tier-aware questionnaires | Manual, often a single template | 181Q, 32Q and 16Q, routed by tier |
| AI cross-checks answers against the vendor's own evidence | Takes the answer as truth | Reads the evidence and cites what it read |
| Drift detection on resubmit | Point-in-time only | Opens a finding on regression |
| Attack-chain narrative per vendor | Risk register only | MITRE-aligned, written for a board |
| Multi-perspective AI review | Single lens, or none | Four specialists and a synthesiser |
| Regulator evidence pack (DORA Art.28, NIS2 Art.21, FCA SYSC 13, UK GDPR Art.28, ISO 27001:2022 A.5.19–5.22, NIST CSF supply chain) | A paid add-on | Six packs, included, with gaps shown as gaps |
| Senior-leader approval by public link | Login required | Signed link, works on a phone |
| Concentration map, where your vendors share infrastructure | Manual analysis | Provider, region, country and type |
Sector ready
They care whether you can answer one question: how do you know what your vendors are doing? Penthropic gives you an answer you can defend, and shows a gap as a gap.
Questions on outsourcing oversight, operational resilience and ICT third-party risk carry the reference they were written against. When an answer regresses and a finding is raised, that finding inherits the question's reference. Findings raised by external monitoring or by the attack-surface scan carry no reference at all, and show none rather than a guess.
NIS2 Art.21 covers third-party risk management as a mandatory control. Penthropic ships a NIS2 evidence pack that groups your vendor's answers and uploaded documents under seven named Art.21 measures. A measure nothing evidences is shown as a gap, and one the evidence cannot grade is shown as unassessable, never folded into covered. We do not tag findings to an article paragraph, because the question bank does not carry NIS2 references and we will not print one we did not map.
Processor obligations, international transfers, breach notification, retention and data subject rights. The bank carries a UK GDPR or GDPR article reference on 34 questions. The ICO Art.28 evidence pack is a separate view and is not built from that reference: it matches a keyword against each question and its references, and gathers answers, vault documents and sub-processor records under six named Article 28 obligations. That draws in 21 questions from the bank, 8 of them from those 34. An obligation nothing matches is shown as a gap.
NIST CSF is the densest mapping in the bank: 60 questions carry a NIST CSF reference, 15 carry an ISO 27001 reference and 5 name SOC 2 at the report level. In total 133 of the 422 built-in questions carry a control reference. The rest carry none, and the report shows none, because a mapping we did not make is not a mapping.
Scorecard methodology
The scorecard pairs passive asset discovery with continuous threat-intelligence checks. Every signal is observable from outside the vendor, and a check that could not run is recorded as unchecked rather than as clean.
We map each vendor's external attack surface using Certificate Transparency logs (crt.sh), the Wayback CDX API and subdomain enumeration via Cloudflare DNS over HTTPS. We then resolve IPs, detect cloud provider ranges (AWS, GCP, Azure, Cloudflare, Fastly), and fingerprint the tech stack from HTTP response headers. The collector that found each asset is recorded against it, so the provenance travels with the asset.
After mapping assets we run lightweight checks: DMARC policy (published, monitor-only or enforced), SPF record presence and a missing all qualifier, MTA-STS policy presence, DNSSEC presence, TLS certificate expiry, subdomain takeover (dangling CNAME), registered typosquat domains, missing HSTS, CSP and clickjacking headers, publicly catalogued breaches of the vendor's own domain in Have I Been Pwned, and domain reputation from AlienVault OTX. Each check maps to a severity score from 1 (low) to 8 (critical). When a lookup fails, the DNS, breach and reputation checks record an explicit unchecked result with the reason, rather than reporting the vendor as clean.
Not on the scorecard. Legacy TLS version detection is not built for vendor scans and does not run. Ransomware leak-site matching is built and sweeps daily, but a hit lands in Dark Web Watch as a match for review, not as a scorecard finding, so it does not move a vendor grade. VirusTotal and Google Safe Browsing left the vendor scan on 17 September 2026, because both licence their free tiers for non-commercial use only. Their rows stay on the vendor record, marked not checked with that reason, rather than disappearing.
No active port scanning, no exploit probing, no destructive tests. We perform read-only passive reconnaissance, we observe what is already publicly visible on the internet. We do not access dark web marketplaces, purchase stolen credentials, or enumerate internal systems. Requests that reach your vendor name us in the User-Agent, so they can identify our traffic in their logs: Penthropic-Scorecard/1.0 from the asset-discovery collectors, and Penthropic-Signal/1.0 from the security-header and lookalike-domain checks.
Each finding is assigned a severity from 1 (Low) to 8 (Critical). The scorecard score is 100 − ∑(active finding weights), clamped to 0–100, then converted to a letter grade: A (90+), B (80–89), C (70–79), D (60–69), F (<60). Severity bumps apply when a scan finding directly contradicts a security claim your vendor submitted in their questionnaire.
Every finding can be disputed. Vendors log in to the vendor portal and submit a written explanation or evidence. The dispute is recorded against the finding with the vendor's reason and the date, and the finding is marked as awaiting review. Raising a dispute does not move the score; the decision on it does. If we uphold the dispute we are agreeing the finding was wrong about the vendor, so it is marked resolved and removed from the active score, with the decision, the written reason and the reviewer recorded against it. If we reject it, the finding stands exactly as it was. A finding also leaves the active score when it is closed or when the risk is formally accepted, and each of those is a recorded action with an owner. Vendors can also request a rescan after remediating an issue.
Asset graphs are rebuilt weekly (Sunday, 03:00 UTC). Threat-intelligence signals refresh nightly. Score snapshots are written after each sweep so trend data is always current. Questions about a specific finding? Contact us at security@penthropic.ai or visit penthropic.ai.
Next step
Thirty minutes on one of your own vendors: what the AI would read, what it would check the answers against, and who on your side would see the result.
TPRM is part of the platform subscription. There is no separate TPRM price list. Pricing is here, and the vendor allowance in each tier is stated on it.
Six regulator evidence packs included. London based, UK hosted. Reply within 1 business day.