The claim
On 18 November Strike Graph launched Integration AI. It builds a connector to any software that exposes an API: one agent reads the vendor’s public API documentation and drafts a specification, and a second agent builds the integration (Strike Graph, vendor, Business Wire release). The connector then collects compliance evidence, and Strike Graph says that evidence counts toward every framework a customer pursues, more than 30 of them, including SOC 2, ISO 27001, HIPAA, and CMMC (Strike Graph, vendor, Business Wire release).
The pitch is speed. Building a traditional integration previously took two to three weeks (Strike Graph, vendor, as relayed by Security Boulevard). Strike Graph also says hundreds of organisations have cut compliance timelines by more than 86% (Strike Graph, vendor, Business Wire release). No method, baseline, or sample comes with that figure.
So the claim under test is narrower than the launch: a connector written by AI can be the source of evidence an auditor relies on. The software is not the hard part. The sign-off is.
What Strike Graph says it built in
To its credit, the release lists its safeguards in plain terms (Strike Graph, vendor, Business Wire release):
- Read-only by default. Endpoint allowlists are checked against the vendor’s documentation.
- No credentials in the build. The building agent has no access to customer credentials, and a specification that contains credentials is rejected.
- A separate publish step. Publishing is locked down behind an approver allowlist and a kill switch.
- People in the loop. Humans review the research plan, the code, and the production deployment.
Every one of these is the company’s own statement. The release was distributed through Business Wire; we read it through its republication on FinancialContent. Security Boulevard covered the launch, but relays the company’s claims with no customer, analyst, or auditor voice. Dark Reading also ran it; we could not read that page, and it appears to be the same release. We found no customer, no auditor, and no independent test.
The test: seven questions an auditor will ask
We put the questions an auditor asks of any system that produces evidence to what Strike Graph has published. A question counts as answered only if the release or the coverage states an answer. A question counts as shown only if someone outside Strike Graph has demonstrated it.
| Auditor’s question | Strike Graph’s stated answer | Shown independently? |
|---|---|---|
| What can the connector touch? | Read-only by default, endpoints allowlisted against vendor docs | No |
| Who holds the credentials? | Not the build agent; specs with credentials rejected | No |
| Who reviewed the code that produces the evidence? | Humans review the plan, the code, and the deployment | No |
| Who can publish a new connector? | A locked-down step behind an approver allowlist | No |
| How do you stop a bad connector? | A kill switch | No |
| Where is the review record, and can an auditor see it? | Not stated | No |
| Who can switch a connector from read-only to write? | Not stated. “By default” implies someone can | No |
| Total | 5 of 7 stated | 0 of 7 |
The counts are our maths on the release and the Security Boulevard report. Five stated answers is a better start than most AI launches give. Zero shown is where every AI launch starts. The two unstated rows are the ones an auditor will open with, because they decide whether the review can be evidenced at all.
Where this breaks
- Every source is the company. The release, its republication, and the Security Boulevard account all trace to Strike Graph. Nothing here is corroborated.
- The 86% has no method. No baseline, no sample, no definition of a compliance timeline. Do not plan an audit calendar on it.
- Reviewed once is not reviewed forever. A connector that reads an API today reads a changed API next quarter. The release says nothing about re-review when the source system or the connector changes.
- Evidence reuse cuts both ways. If one connector’s evidence counts toward 30 frameworks, one bad connector fails 30 frameworks.
- We have not tested it. Nothing here is measured by Prestanda.
Building the connector is Strike Graph’s job. Proving to your auditor that it reads what it claims is yours.
Five steps before AI-built connectors feed your audit
- Step 1. Ask your auditor first. Before you switch anything on, ask whether they will accept evidence from a connector your GRC vendor’s AI wrote, and what they need to see. Owner: compliance lead.
- Step 2. Get the two missing answers in writing. Where the review record for each connector lives, who can read it, and who can switch a connector out of read-only. Owner: GRC lead with procurement.
- Step 3. Give connectors their own read-only identity. A dedicated service account per source system, scoped to the objects the evidence needs, never a person’s login. Owner: the system owner for each source.
- Step 4. Pilot on one low-risk control. Run an AI-built connector beside your current evidence for one audit period and compare the two. Owner: compliance lead.
- Step 5. Set a re-review trigger. Re-review a connector whenever the source API version or the connector code changes, and keep that record with the evidence. Owner: GRC lead.
Missing
- A named customer or auditor who has relied on this evidence.
- Any independent test of Integration AI.
- The method behind the 86% figure.
- How the human review is recorded, and whether a customer or auditor can see it.
- Who can change a connector from read-only, and how that change is logged.
- What triggers a re-review when the source system changes.
- Pricing.
Frequently asked questions
What is Strike Graph Integration AI?
A feature Strike Graph launched on 18 November 2026 that builds integrations to any software with an API. One agent reads the vendor's public API documentation and drafts a specification; a second agent builds the integration, which then collects compliance evidence. That is Strike Graph's own description in its Business Wire release; no independent test has been published.
Can an auditor rely on evidence collected by an AI-built connector?
Only if you can show who reviewed the connector, what it can read, and that it has not changed since the review. Strike Graph says humans review the research plan, the code, and the production deployment, and that connectors are read-only by default. It has not published how a customer or an auditor sees that review record.
Is the 86% reduction in compliance timelines verified?
No. Strike Graph says hundreds of organisations have reduced compliance timelines by more than 86%, but gives no method, baseline, or sample. Treat it as a marketing claim, not a planning figure.
Living page. Last updated 19 Nov 2026. We update this note when Strike Graph publishes how the review record is kept, when a customer or auditor speaks on the record, or when an independent test appears. Want a second opinion on what your auditor will accept from automated evidence? Talk to us.