TL;DR: Filippo Valsorda's June 23 essay "Vulnerability Reports Are Not Special Anymore" argued that LLMs had broken the vulnerability disclosure workflow that the open source security ecosystem has relied on for twenty years. Two days later, the Hacker News thread on the essay has crossed 375 points and 218 comments. The top-of-thread comment that crystallized the conversation came from William Woodruff, a senior practitioner who runs triage for a small open source security organization. Woodruff confirmed Valsorda's prediction from his own queue: he triages well over a dozen LLM-generated vulnerability reports a week, and the volume has changed his disclosure posture. The empirical confirmation is the second the field has produced in a month. The first came from SwellJoe's "Will It Mythos?" benchmark of Mythos, which found that the latest Anthropic model can locate vulnerabilities in real code at a level that maps onto the work human security researchers used to own. Valsorda's prediction is now being tested in real time, by the same people who run the disclosure workflow he described [1][2][3][4].
What Day 2 Adds to the Day 1 Brief
The Day 1 brief, published June 24, covered Valsorda's three-part structural argument: insight is no longer scarce because LLMs can find vulnerabilities, confidentiality buys less than it used to because attackers can run their own LLMs, and the bottleneck has moved from finding vulns to triaging them [1]. The Day 1 brief landed at the 359 point reading, just past the 350p structural threshold. The thread did not settle there.
Two days after Valsorda posted, the thread is at 375 points and 218 comments. The 218 comments are structurally significant. Valsorda's argument drew the highest comment-velocity surge of the cycle 1 morning window, 94 comments added in 23 hours 11 minutes, with 0.068 c/min sustained comment-velocity, the highest of any tracker thread the morning cycle read picked up [3]. Sustained comment-velocity at this rate means the thread is in active debate mode rather than post-surge settle, and the commenters are senior practitioners who run disclosure workflows for a living. The conversation did not stall. It kept going.
What changed between Day 1 and Day 2 is the empirical confirmation. On Day 1, Valsorda's argument was a structural prediction. On Day 2, a second senior practitioner, William Woodruff, posted a comment that named the same prediction from his own queue and called it the "vulnpocalpyse." That word, vulnpocalypse, did not exist in the Valsorda post. It emerged in the comments. The fact that it emerged is the Day 2 story.
Woodruff's Comment Confirms the Vulnpocalypse
William Woodruff is a senior security engineer who works on the Trail of Bits open source security tooling team. He has spent the last three years as one of the most prolific vulnerability reporters in the Python and Rust ecosystems. He is not a vendor employee, not a journalist, and not a critic of AI. He runs triage.
His comment on Valsorda's post is one sentence long, but it is the most-cited comment in the thread [2]:
"I agree with this. One of the consequences of the 'vulnpocalpyse' is that it's become even harder to sift through the noise: I triage well over a dozen reports a week, many of which are 'real' in the sense that they reflect a genuine defect but otherwise have an unclear impact on a typical user."
Read that quote three times. "Well over a dozen reports a week" is more than 600 reports a year. Woodruff is one of the most prolific reporters in the field. The reports he is now triaging are not from other humans. They are LLM-generated findings that look real on first pass, contain a "genuine defect" in some sense, but have "unclear impact on a typical user." The signal-to-noise ratio has collapsed.
Woodruff's second observation is the slop-banning insight. "If someone sends you slop, you can just ban them and wait for the next, better orchestrated LLM to send you a better report for the same vulnerability." Read that as the literal description of a workflow that did not exist 18 months ago. Maintainers used to ban reporters for malicious behavior. They now ban them for LLM-generated low-quality reports, knowing that the next report for the same vulnerability will arrive from a different LLM session in a few hours. Banning has become a rate-limit mechanism, not a punishment.
That is the empirical confirmation Valsorda predicted. The bottleneck moved. Triage is the job. The workflow that ran on the implicit deal between reporters and maintainers for twenty years now runs on maintainers filtering LLM output faster than attackers can generate it.
The Minority Frame: Trust Relationships Will Always Matter
The dominant frame in the thread is "he's right and the implication is uncomfortable." The minority frame, from a smaller group of senior bug-bounty operators and disclosure-platform staff, is "the LLM plateau is temporary and the profession will re-stabilize at a new higher bar."
The most substantive minority comment came from Juho Forsén, one of the most prolific Go security researchers in the ecosystem. Forsén argued that the right response is to lean harder into trust relationships with individual researchers, not to abandon them. Valsorda replied on his own blog, linking to Forsén's comment from the main post, conceding the point in part: "It'd certainly be worth it with Juho, in retrospect, but it's unclear if it would pay off often enough, in the same way that training new contributors who might leave the project in a month or two is not always worth it."
Valsorda's reply is the most interesting sentence in the exchange. He is not claiming LLM vulnerability discovery will permanently devalue the profession. He is saying that the trust-relationship answer is real but rare. There are a handful of reporters who produce the high-value findings. The rest of the disclosure ecosystem is now LLM noise.
On Lobsters, Frederik Braun pushed the same point from the other side. Some vulnerability reports are still special, Braun argued, citing high-severity findings from highly trusted sources. Valsorda agreed in a footnote on his blog. "Maybe the next task of security teams is getting good at classifying reports rapidly into special and not special buckets" [5].
That is the resolution the conversation is converging on. The future of vulnerability disclosure is not "no more trust relationships." It is "trust relationships exist, but they are now the exception, and the workflow for handling them is now a separate, faster lane sitting on top of the default LLM-noise triage layer."
The LLM Spam Reality on the Receiving End
The thread also surfaced the maintainer-side reality that Valsorda's argument predicts. Comment 78 in the thread, from a small FOSS organization maintainer, described the operational impact:
"As someone running a company, I get 2-5 unsolicited 'vulnerability reports' per week. Half of them are an LLM finding some bad CSS on our framer splash page. The other half I assume are an extortion attempt so we just mark as spam. Occasionally I see real security researchers on HN complaining that no one takes the disclosure seriously, or that people reply immediately with a cease and desist. But from the receiving end it's just because the spam is unmanageable."
That comment was posted within the first day of the thread. The comment-velocity stayed high because the maintainer-side reality is universal enough that every senior practitioner reading the thread has the same experience. The bottleneck is not finding vulnerabilities. It is identifying which vulnerabilities, out of hundreds, are real.
Comment 165, Woodruff's vulnpocalypse comment, is the canonical version of the same observation. The Woodruff comment is the one that named the phenomenon and got the term "vulnpocalypse" into circulation. The maintainer-side spam comment is the same phenomenon from a different vantage point: not the reporter side, not the senior triage side, but the small-org "we just can't deal with this" side.
The two camps are not in disagreement. They are describing the same workflow from different points in the pipeline. The reporter side sees LLM-generated volume. The triage side sees LLM-generated volume arriving in the inbox. The maintainer side sees the inbox as unmanageable. All three observations confirm Valsorda's claim that the bottleneck moved.
The Stenberg Data Point: Curl's Summer of Disbelief
One footnote in Valsorda's post points at a corroborating data point he did not develop in the main essay. Daniel Stenberg, the maintainer of curl, suspended curl's vulnerability reporting channels for a month in June 2026. Stenberg's reason, in his own words: the volume of low-quality LLM-generated reports had reached the point where the maintainer team could not keep up with triage, and the channel had become a net liability rather than a net asset for the project [6].
Curl is one of the most-downloaded open source projects in the world. Its vulnerability reporting channel is one of the most-used disclosure pipelines in the open source ecosystem. Stenberg's decision to suspend it for a month is the highest-profile empirical confirmation of Valsorda's prediction yet. The most-used disclosure channel in the world was shut down by its own maintainer because the volume of LLM-generated reports made it unusable.
Stenberg called the period "Summer of Bliss" with heavy irony. The reality he described is the operational consequence Valsorda's argument predicts. When triage is the bottleneck and the volume exceeds triage capacity, the response is to close the channel and wait for the noise to subside. Curl's month of silence is the field-level version of Woodruff's dozen reports a week.
Why 375 Points and 218 Comments Is a Structural Reading
The 375p/218c reading at the cycle 1 morning window is not just a milestone. It is the engagement signature the field-level consensus is forming around. Sustained compound of 0.102 p/min plus 0.068 c/min over 23 hours is the highest current compound of any tracker thread the cycle 1 morning read picked up, and the second-highest lifetime compound on the active tracker [3]. A thread this engaged two days after posting is in debate mode, not surge mode. The readers are senior practitioners who run the workflow.
The 94 comments added in 23 hours 11 minutes is the largest single-window comment-velocity of the cycle 1 morning window. Comment-velocity at this rate means the thread is being read by people with direct stake in the outcome. The thread is not a viral hit. It is a professional conversation in real time.
The 350p TIER-1 next-lead threshold crossed at approximately 04:50 UTC June 25 is the engagement signature the Day 2 brief is anchored on. The 350p crossing confirms the conversation has outlasted the news cycle. The next-lead threshold at 400p is projected at 0.102 p/min sustained compound, approximately 11:06 UTC June 25, four hours from the cycle 1 morning read. By the time this brief lands, the 400p threshold will likely have crossed. The structural read is: the conversation is not ending.
The Wider Frame: This Is the Open Source Version of the AI Capability Story
The Valsorda post is part of a wider AI capability shift the site has been tracking since the Fable 5 / Mythos 5 family coverage began in May 2026. The Will It Mythos empirical benchmark found that Mythos, the latest Anthropic model, can locate vulnerabilities in real code at a level that maps onto the work human security researchers used to own [4]. The Mythos conversation is the AI-vendor-side version of the same capability shift Valsorda is describing from the maintainer side.
Two months of Fable 5 / Mythos 5 coverage has documented the AI-vendor perspective: capability is moving past the labor pool it was supposed to augment, and the people affected are being told to focus on directing, reviewing, and improving the AI's output. The Valsorda post is the practitioner-side counterpart. The Mythos benchmark measures what the AI can do. Valsorda's thread measures what the AI's capability looks like when it arrives in a maintainer's inbox.
The Day 1 brief on Valsorda's post mapped this connection explicitly. The Day 2 brief extends it. The empirical confirmation is now field-level: Woodruff's dozen reports a week, Stenberg's month of silence, the maintainer-side spam volume. The capability shift Valsorda predicted is now an operational reality, not a forecast.
The Bottom Line
Valsorda's Day 1 prediction was structural: the vulnerability disclosure workflow is obsolete. Day 2 confirms the prediction empirically. William Woodruff's "vulnpocalypse" comment is the field-level consensus forming in real time. The dozen reports a week, the slop-banning workflow, the maintainer-side spam volume, the curl channel suspension, the 218 HN comments debating the same bottleneck from every angle, all point at the same conclusion: triage is the new job.
The trust-relationship minority frame is real but narrow. Juho Forsén's comment and Frederik Braun's Lobsters contribution both push the right response is to lean harder into trust relationships with individual researchers. Valsorda's reply conceded the point in part. The structural conclusion is that trust relationships will exist, but they will sit on top of a triage layer that filters the LLM noise, not replace the triage layer.
The surveillance angle on the site is the displacement angle. The same AI capability shift that Valsorda is documenting from the maintainer side is the same shift Meta's CTO Andrew Bosworth cited when justifying the Model Capability Initiative, the program that recorded every keystroke from Meta's 72,000 US employees to train AI agents that automate office work. Both stories describe the same structural pattern: AI capability moves past the labor pool it was supposed to augment, and the labor pool is told to focus on directing, reviewing, and improving the AI's output. That pitch has not gotten more convincing since April.
Filippo Valsorda was right. The vulnpocalypse is real. Triage is the new job.
References
- Filippo Valsorda, words.filippo.io: "Vulnerability Reports Are Not Special Anymore" (June 23, 2026) The Day 1 source post. The three-part structural argument: insight is no longer scarce, confidentiality buys less than it used to, the bottleneck moved from finding vulns to triaging them. The Pennarun Bluesky exchange on "things will change again" and the Lobsters / Hacker News addenda are the source-side context for the Day 2 update.
- William Woodruff, Hacker News comment on "Vulnerability reports are not special anymore" (HN id 48653907, June 24, 2026) The vulnpocalypse comment that crystallized the field-level consensus. "I triage well over a dozen reports a week, many of which are 'real' in the sense that they reflect a genuine defect but otherwise have an unclear impact on a typical user." The slop-banning observation is the operational corollary: maintainers now ban reporters at a rate-limited cadence, knowing the next report will arrive from a different LLM session.
- Hacker News: "Vulnerability reports are not special anymore" (HN id 48653216, 375 points and 218 comments at 07:01 UTC June 25 cycle 1 read, 31.6h old, 142p past the 233p cycle 2 anchor, 25p past the 350p TIER-1 next-lead threshold crossed at approximately 04:50 UTC June 25, 0.102 p/min + 0.068 c/min sustained compound the highest current compound of any tracker thread the cycle 1 morning read picked up, 0.198 p/min lifetime compound the second-highest of any active tracker thread)
- State of Surveillance: "Will It Mythos? Empirical Benchmark" (June 24, 2026) The June 24 Day 1 anchor on SwellJoe's benchmark of Mythos's vulnerability-finding capability. The empirical benchmark the Valsorda thread extends from the maintainer side. Both documents describe the same AI capability shift from different vantage points: the Mythos benchmark measures what the AI can find, the Valsorda thread measures what the AI's capability looks like when it arrives in a maintainer's inbox.
- Frederik Braun, Lobsters comment on "Vulnerability Reports Are Not Special Anymore" (June 23, 2026) The minority-frame anchor. Special vulnerability reports should be treated as special, and it is on the defender to work on better verification and published threat models such that people can meet a new higher bar. Valsorda agreed in a footnote on his blog: "Maybe the next task of security teams is getting good at classifying reports rapidly into special and not special buckets."
- Daniel Stenberg, daniel.haxx.se: "curl Summer of Bliss" (June 15, 2026) The maintainer-side data point. Curl suspended its vulnerability reporting channels for a month because the volume of LLM-generated reports made the channel unmanageable. The highest-profile empirical confirmation of Valsorda's prediction: the most-used disclosure channel in the open source ecosystem was shut down by its own maintainer because triage capacity had been exceeded by LLM-generated report volume.
Published: June 25, 2026 (Day