This page requires JavaScript to display.

AI-found PHP flaws

AI & tech · Heat 9.7 / 10 · 14 shows · 3h 22m airtime

On October 22, Wiz disclosed severe PHP vulnerabilities found using AI, in the language behind an estimated 70 percent of websites. Governments issued alerts urging organizations to patch or take sites down, and pages at Wikipedia, the BBC and several governments went dark. The discussion has moved from the find itself to the speed of exploitation and the slow response: knowledge spread about 10 hours ahead of disclosure, exploit attempts went from thousands to millions, and patches were still rolling out 48 hours later. Only 30 to 40 percent of websites are expected to be patched by Monday, while the PHP Foundation says its patch took over a month because it is under-resourced.

How 14 podcasts covered it

  1. Global News Podcast, 2026-10-22 (2:26 to 5:57)
    Reports the PHP flaws and government alerts
    Reports the AI-discovered PHP flaws and the wave of government alerts and site shutdowns, with Wiz calling them the most severe vulnerabilities AI has found. Frames the story around AI reshaping both attack and defense.
    New angle: Wiz says the AI-found PHP flaws are the most severe vulnerabilities ever discovered by AI.
  2. Global News Podcast, 2026-10-22 (5:18 to 8:02)
    Interviews the correspondent on the leaked PHP flaw
    Interviews the cyber correspondent on the leaked PHP flaw: AI found the exploit chain humans had parked as theoretical, humans then leaked it, and AI on the other side is shortening the lag before working attacks.
    New angle: AI has shortened the lag between a vulnerability being public and a working exploit existing.
  3. Breaking Points with Krystal and Saagar, 2026-10-22 (41:15 to 47:01)
    Walks through the PHP flaw and nearby breaches
    Walks through the Wall Street Journal report on the PHP bug in plain terms and lists the surrounding incidents from the past 48 hours, from the California student data theft to the Zboard breach. The framing is that AI has handed attackers superhuman capability.
    New angle: AI tools found a critical PHP vulnerability that could let attackers take control of any system running it.
  4. Breaking Points with Krystal and Saagar, 2026-10-22 (47:04 to 55:10)
    Adds banks banning AI agent access
    Extends the PHP disclosure into the exploit side: the bug was known a decade ago and judged too hard to exploit, and the same AI tools are now available to ordinary users. Adds that JPMorgan Chase and a few other banks have banned AI agent access to their systems.
    New angle: JPMorgan Chase and a few other banks have banned any AI agent access to their systems.
  5. Bloomberg Tech, 2026-10-22 (33:26 to 41:22)
    Interviews Brockman and Prince on the flaws
    Interviews Brockman and Prince on the AI-found PHP flaws. Brockman says OpenAI's internal model autonomously found the full exploit in September, a month after Calif's guided find; Prince says Cloudflare is seeing millions of attempted attacks per minute and worked with WordPress and the PHP community before disclosure.
    New angle: OpenAI's internal model autonomously found the full PHP exploit in September, a month after Calif's guided find.
  6. Bloomberg Tech, 2026-10-22 (41:19 to 44:06)
    Floats authorized defenders patching systems
    Brockman, on the record, floats a policy idea rather than a development: authorized defenders scanning the internet and remotely patching vulnerable systems, with the FBI's botnet takedowns as precedent. He also puts a figure on the disclosure - two months of manual process and calls with partners and government.
    New angle: Brockman proposes authorized defenders scan and remotely patch vulnerable systems, citing FBI botnet takedowns as precedent.
  7. The Information's TITV, 2026-10-22 (15:58 to 21:24)
    Walks the patch timeline and the funding gap
    Reporter Aaron Holmes walks through the PHP flaw's timeline: discovery in August, OpenAI reporting the same day as Calif, and a patch that took over a month because the PHP Foundation says it is under-resourced. Adds the foundation's own explanation for the delay.
    New angle: The PHP Foundation says its patch took more than a month because it is under-resourced.
  8. The Information's TITV, 2026-10-22 (21:22 to 31:01)
    Reports the briefings to CISA and the NSA
    Brings on the reporter behind the defender's-window column and reports that OpenAI and other groups briefed CISA, the NSA and the Superintelligence Task Force, including Emil Michael and Jay Clayton, on the PHP flaw late last week before the response was mobilized this week.
    New angle: OpenAI and other groups briefed CISA, the NSA and the Superintelligence Task Force on the PHP flaw late last week.
  9. The AI Daily Brief, 2026-10-22 (14:47 to 16:48)
    Adds the ten-hour gap before disclosure
    Retells the PHP 2 Shell disclosure with the Heartbleed comparison and OpenAI's defender's-window blog post, then adds one detail the day's coverage had not put on the record: exploit knowledge circulated about ten hours before the patch went public.
    New angle: Knowledge of the PHP exploit spread about 10 hours ahead of disclosure, before servers could be updated.
  10. The AI Daily Brief, 2026-10-22 (16:53 to 19:22)
    Adds Prince's exploit attempt timeline
    Collects on-the-record reactions from founders and engineers to the PHP 2 Shell disclosure, then pushes back on rationalists claiming the exploit as their predicted risk. The new material is Matthew Prince's attempt-count timeline across the Cloudflare network.
    New angle: Prince says exploit attempts went from thousands pre-announcement to tens of thousands in 45 minutes, then millions.
  11. The AI Daily Brief, 2026-10-22 (26:23 to 27:51)
    Recaps the internet security proposal and Casado's warning
    Reads Brockman's 'Towards Securing the Internet' post aloud - authorized government scanning and remote patching, a new phase of internet-scale vulnerability discovery - and pairs it with Casado's warning that things get much worse before better. A recap of the proposal already on the record, closing the episode on the host's long-term optimism.
  12. TBPN, 2026-10-22 (0:19 to 28:58)
    Recaps the PHP2Shell timeline from Cloudflare
    Matthew Prince walks through the PHP2Shell timeline from Cloudflare's side: the quiet two-month coordination with OpenAI, Wiz, Automattic and the PHP Foundation, the free WAF mitigation, and the attack surge after the patch. Mostly retells what is already on the record, plus a detour into weak root-key-server passwords.
  13. Equity, 2026-10-23 (2:51 to 16:49)
    Recaps the PHP disclosure and link-rot worry
    Recaps the PHP disclosure at length: the decades-old flaw, LLMs at OpenAI and Wiz finding it, governments and sites going dark, and Prince's escalation numbers. It stays inside the week's coverage, adding only the link-rot worry and startup speculation.
  14. The Vergecast, 2026-10-23 (36:58 to 46:47)
    Walks through the PHP flaw and the embargo break
    Walks through the PHP flaw in plain terms: the memory-exhaustion bug, the 70 percent of the web on PHP, the prompt that found it, and the embargo break by Chain Guard. It retells the disclosure and the fallout without adding anything new to the record.
  15. The Vergecast, 2026-10-23 (46:47 to 50:44)
    Argues autonomous defense is the only answer
    Ties the PHP disclosure to the Gates interview and the Hugging Face hack, arguing the only real answer is autonomous defensive AI and that some sites had to go offline. Reaction and framing on top of the week's reporting, with no new development.
  16. Breaking Points with Krystal and Saagar, 2026-10-23 (53:45 to 1:01:45)
    Frames the sell-off and cyber incidents as one
    Frames the sell-off and the cyber incidents as one story: AI models now find and exploit vulnerabilities at expert level, with a PHP flaw touching an estimated 70 percent of websites and AI agents inside bank accounts. Mostly alarm and argument, with no new document beyond the incidents themselves.
    New angle: A PHP flaw found with AI is estimated to affect 70 percent of websites worldwide.
  17. Breaking Points with Krystal and Saagar, 2026-10-23 (1:01:45 to 1:15:57)
    Explains how AI shrinks the patch window
    Explains how AI collapses the window between patch release and exploitation, using the PHP flaw as the case study. Reads an IT worker's account of servers nobody can locate and an Anthropic employee's warning that the next four days get worse, then pivots to AI-named layoffs at Amex, Barclays, eBay and Netflix.
    New angle: An Anthropic employee, Jason D Clinton, says the next four days get worse and security leaders should ask boards for help now.
  18. Breaking Points with Krystal and Saagar, 2026-10-23 (1:15:57 to 1:22:31)
    Debates whether AI defenses outrun attackers
    Debates whether AI defenses can outrun AI attackers, using the Praetorian guard analogy, and argues the companies created the problem to sell the solution. It stays inside the Black October frame, adding only commentary and the Hugging Face detail already on the record.
  19. Daily Tech News Show, 2026-10-23 (22:47 to 25:36)
    Theorizes how the PHP leak spread
    Offers a theory of how the leak spread - a vague mention of a big PHP project reaching someone who pointed an LLM at the code - and cites ChainGuard's own account: it asked Claude to narrow down the two flaws but could not patch them without the time Calif and Wiz spent.
    New angle: ChainGuard asked Claude to narrow down the two PHP flaws but could not build the patch itself.
  20. The Information's TITV, 2026-10-23 (0:59 to 9:12)
    Frames the PHP scramble as credential hygiene
    Interviews 1Password's CEO about the PHP scramble two days in. He says patches are still rolling out and frames the week's incidents as a credential-hygiene problem, with standing access the ticking bomb as the cost to attack falls toward zero.
    New angle: Patches for the mass PHP vulnerability were still being rolled out about 48 hours after disclosure.
  21. Closing Bell, 2026-10-23 (18:31 to 20:35)
    Reports Kurtz's warning on AI attacks
    Covers the PHP 2-Shell rally through Kurtz's warnings: AI attacks doubling monthly, unpatched legacy PHP appliances, and low-skilled groups buying nation-state-grade attacks for the cost of tokens. The hosts compare it to Y2K and tie it to AI-trade negativity.
    New angle: Kurtz says low-skilled groups can now run nation-state-level attacks for the cost of tokens.
  22. Big Technology Podcast, 2026-10-23 (33:07 to 36:03)
    Sets up the Stamos interview with the WSJ disclosure
    Sets up the Alex Stamos interview by recapping the PHP disclosure from the Wall Street Journal: a critical flaw in software behind as much as 70 percent of websites that could let attackers take control. Stamos is introduced as the guest; the substance is a retelling of what earlier coverage already established.
  23. Big Technology Podcast, 2026-10-23 (35:54 to 40:20)
    Explains the bug's pre-auth RCE and patch scope
    Stamos walks through how the PHP bug was found and why it bites: pre-authentication remote code execution, a patch only for the latest PHP family, and companies forced to refactor. The patch-scope point is already on the record; the rest is explanation.
  24. Big Technology Podcast, 2026-10-23 (40:20 to 44:54)
    Walks through who exploits the PHP bug
    Stamos walks through who is exploiting the PHP bug and why: Russian ransomware crews, Chinese access brokers and Iranian groups, with attackers patching behind themselves to lock out rivals. He adds that a Chinese group hit South Korean banks with fully automated AI attack chains.
    New angle: A Chinese group broke into a number of South Korean banks using fully automated AI attack chains.
  25. Big Technology Podcast, 2026-10-23 (44:54 to 47:50)
    Relays Stamos's forecast of breach notifications
    Recaps Prince's Cloudflare figures on the exploitation surge, then relays Stamos's forecast of breach notifications, secondary breaches from footholds and password dumps sold repeatedly. The numbers are already on the record; the forward look is prediction.
  26. Big Technology Podcast, 2026-10-23 (1:00:13 to 1:02:35)
    Argues for rewriting software in memory-safe languages
    Stamos closes the episode on the PHP bug's decade of dormancy and the case for rewriting internet software in memory-safe languages, arguing AI makes the refactor feasible now. Opinion and prescription rather than new reporting; it stays inside the week's PHP coverage.
  27. Up First, 2026-10-24 (1:58 to 6:39)
    Explains the PHP flaw and AI discovery wave
    Explains the PHP flaw and the AI-driven wave of vulnerability discovery behind it, with practical advice to rotate passwords and patch software. It frames the moment as Black October without adding a specific new document or figure.
    New angle: Internet security researchers discovered a vulnerability in PHP two months ago.
  28. The Best One Yet, 2026-10-26 (9:14 to 10:26)
    Reports the NSA's PHP patch warning
    Covers the NSA's PHP warning: the flaw found by an AI-using research firm, the public patch, and the estimate that only 30% to 40% of sites will be patched by Monday. Ties it to the Knicks phishing story as a cybersecurity lesson.
    New angle: Only 30% to 40% of websites are expected to be patched by Monday.
  29. The AI Daily Brief, 2026-10-26 (10:36 to 12:25)
    Relays Prince and Altman warnings
    Picks up the PHP fallout through the on-the-record warnings: Clinton's 18-month horizon and his reply to critics, Prince telling Reuters he knows of more flaws in other critical software, and Altman saying not all bugs can be patched.
    New angle: Prince told Reuters he knows of additional vulnerabilities in other critical software; Altman says not all bugs can be patched.
  30. The AI Daily Brief, 2026-10-26 (15:35 to 18:54)
    Reports OpenAI's offline training plan
    Surveys the reaction to the vulnerability wave - agent swarms for cyber defense, Noah Smith's end-of-the-open-internet thesis - then goes to the labs' response: OpenAI's Friday blog says frontier training resumes in an offline world, and an experiment shows agents in separate micro-VMs talking through the processor cache.
    New angle: OpenAI will run all frontier RL in a quarantined offline world; agents in separate micro-VMs communicated via processor cache.
  31. Equity, 2026-10-28 (1:00 to 2:47)
    Recaps the patch release and its web-wide scale
    Recaps the PHP patch release and the scale of the flaw - WordPress, Drupal, Joomla and MediaWiki, roughly 70 percent of the web - then hands off to Wiz's CTO for a retelling of what was already reported. Stays inside the week's coverage.
  32. Equity, 2026-10-28 (6:05 to 9:31)
    Describes Project Atlas running frontier models in parallel
    Wiz's CTO describes Project Atlas, the in-house harness that runs frontier models from DeepMind, OpenAI and Anthropic in parallel to try many attack paths at once. He says the PHP bugs would not have been found without AI, against Calif's 90-percent-human framing.
    New angle: Wiz's Project Atlas harness runs on DeepMind, OpenAI and Anthropic models at once to try many attack paths in parallel.
  33. Equity, 2026-10-28 (12:10 to 15:13)
    Details the 12-year-old flaw and two-month fix
    Wiz's CTO walks through the disclosure endgame: the flaw was a 12-year-old architectural problem, the fix took roughly two months from detection, and the inner circle swelled to hundreds of people, which he treats as inherent to coordinated disclosure rather than a leak to blame anyone for.
    New angle: Schwartz says the PHP flaw was a 12-year-old architectural flaw and detection to fix took roughly two months.
  34. Equity, 2026-10-28 (15:13 to 18:57)
    Walks through the coordinated disclosure mechanics
    Schwartz walks through the coordinated disclosure mechanics: the PHP Foundation gave Ubuntu, Red Hat and AWS Linux binary builders a few days' head start, and Wiz shared detection rules with cloud and WAF vendors while running silent detections to check for attacks in the wild before release.
    New angle: Ubuntu, Red Hat and AWS Linux got the patch a few days early to build binary releases.
  35. Equity, 2026-10-28 (19:14 to 21:22)
    Explains why the PHP bugs sat dormant
    Schwartz explains why the PHP bugs sat dormant: the PHP team flagged the area as sensitive as early as 2020, but nobody built a reliable remote exploit until AI let researchers try many attack ideas at once. He warns the one-headliner-a-year cadence is ending.
    New angle: The PHP team flagged the vulnerable area as problematic as early as 2020, but no reliable remote exploit existed until AI.
  36. Odd Lots, 2026-10-30 (0:36 to 3:28)
    Introduces the PHP wave and the AI angle
    Opens the episode by framing the PHP vulnerability wave and the AI angle, then introduces Calif's Tai Duong and Bruce Dang. Mostly setup and the hosts' own uncertainty about cybersecurity; the concrete claims are the 70% figure and Calif and Wiz as discoverers.
    New angle: Calif and Wiz uncovered the PHP flaws, and Anthropic's Mythos, now Project Glasswing, went to companies for defense first.
  37. Odd Lots, 2026-10-30 (13:58 to 20:31)
    Details the PHP discovery and the declined fix
    Goes inside the discovery: Calif's Ryan Osteen found the first PHP flaw within a day or two of starting on July 30, and the PHP Foundation had known of the same bug since 2024 but declined to fix it without an exploit. Calif supplied one the next day.
    New angle: The PHP Foundation knew of the flaw since 2024 and declined to fix it because no exploit existed.
Unpacking...