Akrites staff say similar AI models keep sending maintainers the same bug reports
Akrites CTO Christopher Robinson estimates one security pull request can take an engineer up to eight hours to review. The Linux Foundation announced the shared response team in June to consolidate and validate vulnerability reports.
The Engineer · Build desk
Robinson: reviewing one security pull request can take an engineer up to eight hours; anyone with a $20 account is now a researcher; similar models bring a ton of duplicative reports; enterprises aren't necessarily reporting upstream. Guazzelli: models can't necessarily call a false positive.
- cost Engineers Reviewing a single security pull request can take up to eight hours, Robinson estimates, claim 1
- capability Anyone with a $20 account "Everybody with a $20 account is now a researcher," Robinson said, claim 5
- exposure Akrites Getting a ton of duplicative reports as models work similarly and find similar bugs, Robinson said, claim 6
- constraint AI models Can't necessarily call a false positive, not knowing controls applied elsewhere, Guazzelli said, claim 11
- decision Enterprises Better prepared, but not necessarily reporting findings to the upstream project, Robinson said, claim 9
| Who | How | Kind | Claim |
|---|---|---|---|
| Engineers | Reviewing a single security pull request can take up to eight hours, Robinson estimates | cost | 1 |
| Anyone with a $20 account | "Everybody with a $20 account is now a researcher," Robinson said | capability | 5 |
| Akrites | Getting a ton of duplicative reports as models work similarly and find similar bugs, Robinson said | exposure | 6 |
| AI models | Can't necessarily call a false positive, not knowing controls applied elsewhere, Guazzelli said | constraint | 11 |
| Enterprises | Better prepared, but not necessarily reporting findings to the upstream project, Robinson said | decision | 9 |
What happened
- Robinson described the reports as very verbose and highly inaccurate, with inflated severity and too little data for an engineer to write a patch.
- Guazzelli said models cannot reliably call a finding a false positive because they do not see the controls applied elsewhere in a project.
- Enterprises are better prepared, Robinson said, but they are not necessarily reporting what they find to the upstream project.
Why it matters
- cost A $20 account buys the report, but the review time lands on the upstream project that receives it, and every duplicate bills a second maintainer.
- constraint A central team can drop duplicates, but declaring a finding harmless needs a project's own context, so Akrites still depends on upstream maintainers for that call.
- contradiction Report quality is rising while the maintainer queue does not improve, because the better reports come from paid models and may never be filed upstream.
Robinson put the source of the volume in one line. "Everybody with a $20 account is now a researcher," he told The New Stack [5]. The publication's own framing is that AI has lowered the barrier to vulnerability hunting, so almost anyone can probe an open source codebase whatever their security credentials [10].
Duplication is the problem a shared intake can address most directly. Robinson said: "These models all kind of work similarly and they find similar bugs, so we're getting a ton of duplicative reports." [6] Without a shared queue, every maintainer who receives the same finding reviews it separately. Akrites was set up to consolidate reports [2]. In our view, that lets one reviewer look at a finding once. The interview excerpt does not describe how Akrites deduplicates or how many reports it has handled.
Validation is harder. Laura Guazzelli, the Linux Foundation's security architect for AI and agentic systems, said: "The robots don't have the context." [4] She went on: "That's why they can't necessarily call something a false positive, because they don't understand that you have controls that are additionally applied somewhere else." [11]
That limits what a central team can do alone. It can remove duplicates and chase missing detail. A verdict that a flagged line is harmless depends on knowing the controls elsewhere in the project, and that knowledge sits with upstream maintainers, whom Akrites is meant to coordinate fixes with [2].
Report quality is a separate matter. Robinson said quality improved considerably this year, but the gains are not necessarily reaching open source maintainers [7]. His explanation: "The better reports come from the more expensive paid models." [8] He added: "[Enterprises] are better prepared, but they also are not necessarily taking those findings and reporting them to the upstream project." [9]
A shared team sees what gets filed. In our view, findings that stay inside an enterprise do not reduce the queue that Akrites, or any maintainer, has to work through.
Robinson and Guazzelli spoke to The New Stack at Open Source Summit Europe in Prague [12]. Robinson is CTO of Akrites and chief security architect at the OpenSSF [12].
What to watch
- Whether Akrites publishes intake volumes, duplicate rates or time-to-triage, which would show how much review work it removes from maintainers.
- Whether enterprises that run paid models route their findings through Akrites or keep them in-house.
- How Akrites gets project-specific controls from upstream maintainers when it validates a report.
Clarity's read
What the record supports and how the coverage leans. The claims behind it follow.
Reality
- Evidence35
- Adoption50
- Hype gap
- Insufficient
- Incentives60
- Confidence45
Claim ledger
Ranked by verification strength, evidence, and original report placement.
- [1]
Christopher Robinson estimates that reviewing a single security pull request can take an engineer up to eight hours.
- [2]
Akrites, an initiative announced by the Linux Foundation in June, was established to provide a shared security incident response team that consolidates vulnerability reports, validates findings, and coordinates fixes with upstream maintainers.
- [3]
Robinson said AI-generated reports are "very verbose" and "highly inaccurate", that they "overestimate the severity of things", and that they "generally don't give you the right data that an engineer needs to be able to write a patch, or understand what it is."
- [4]
Laura Guazzelli said: "The robots don't have the context."
- [5]
Robinson said: "Everybody with a $20 account is now a researcher."
- [6]
Robinson said: "These models all kind of work similarly and they find similar bugs, so we're getting a ton of duplicative reports."
- [7]
Robinson says the quality of AI-generated vulnerability reports has improved considerably this year, but the benefits are not necessarily reaching open source maintainers.
- [8]
Robinson said: "The better reports come from the more expensive paid models."
- [9]
Robinson said: "[Enterprises] are better prepared, but they also are not necessarily taking those findings and reporting them to the upstream project."
- [10]
AI has lowered the barrier to entry for vulnerability hunting, allowing almost anyone to start probing open source codebases regardless of their security credentials.
- [11]
Guazzelli said: "That's why they can't necessarily call something a false positive, because they don't understand that you have controls that are additionally applied somewhere else."
- [12]
The New Stack interviewed Christopher Robinson, CTO of Akrites and chief security architect at the OpenSSF, and Laura Guazzelli at Open Source Summit Europe in Prague.
Get this ledger inside ChatGPT. Ask about any of our stories and the answer brings its figure, each claim’s verdict and the sources. Free, no account.
Sources
1 independent publisher whose own reporting we read for this story.
Topics and entities
Follow any of these and your For You feed starts watching them — no settings page required.
Topics
- AI-generated vulnerability reportsFollow
- Open Source SecurityFollow
Entities
- AkritesFollow
- Linux FoundationFollow
- OpenSSFFollow
- Christopher RobinsonFollow
- Laura GuazzelliFollow
- The New StackFollow