The apps are only the beginning.

The briefing opens with attempts to exploit two critical PHP vulnerabilities known as PHP2Shell. The flaws permit unauthenticated remote code execution and led to emergency patches on October 21, but researchers have not confirmed a successful compromise in the wild. Sophos observed attackers trying to deploy Potassium, a Mirai botnet variant, although that attack failed. Exploitation appears to require particular server configurations or a way around memory protections, making it more difficult than initially expected. Volncheck developed an exploit that bypasses one protection, and an exposed attacker server contained custom tools plus unsuccessful attempts against dozens of targets. Patching remains difficult: Qualys estimates that about 110 million PHP instances use unsupported versions that will not receive fixes. NowSecure analyzed more than 15,000 Android and iOS applications and found 426 apps connecting to potentially affected PHP servers; 170 connected to servers running outdated PHP versions. The apps themselves are not vulnerable—the risk is in their back-end servers—and the findings do not prove that any server is unpatched or exploitable. One advertising provider was contacted by 151 apps, illustrating the danger of shared third-party infrastructure. NowSecure recommends verifying patching with infrastructure providers, upgrading unsupported PHP, and investigating possible exploitation.

Former CISA director Chris Krebs told CNBC that AI-powered tools are discovering software vulnerabilities at an unprecedented rate. He described the trend as a possible “Black October” for cybersecurity: defenders are finding and patching more flaws, but the discoveries may arrive faster than organizations can respond. He cited PHP2Shell as an example and reiterated that, despite exploitation attempts, no successful compromises had been confirmed. Krebs said roughly 70% of the world's internet sites run on PHP, making the “last mile” the major concern. Large banks and other enterprises generally have the staffing and resources to deploy emergency fixes; Krebs said bank CISOs he spoke with had received early access to PHP patches and were already rolling them out. His greater concern is smaller businesses, rural hospitals, utilities, and managed service providers that lack dedicated security teams. Third-party services can expose whole networks, and legacy systems make patching harder. Krebs expects conditions to get worse before they improve because AI will continue to uncover vulnerabilities faster than organizations can address them.

Cloud development platform Railway confirmed a cyberattack that may have exposed customer passwords, API keys, and other secrets stored in encrypted environment variables. The company believes only a subset of customers was affected, but told all users to revoke and rotate credentials. Railway temporarily suspended new builds and deployments while investigating. It has not identified the number of affected customers, the root cause, or exactly which systems were compromised, and suspects the incident may be part of a broader campaign against AI inference providers. OpenCode, an AI coding tool, reported a similar attack earlier in the week affecting credentials for more than 1% of its customers. Railway said it would publish a postmortem. Hewlett-Packard Enterprise issued fixes for 72 ArubaOS vulnerabilities, including eight maximum-severity flaws that enable remote unauthenticated authentication bypass or code execution with no user interaction. The bugs affect ArubaOS 8 and 10 in enterprise access points, controllers, and gateways. HPE urged immediate patching and recommended segmenting management interfaces and protecting them with firewall controls; it reported no public exploit code or known exploitation. Separately, Quest Diagnostics took its Quantum Lab Services Manager offline after unauthorized access potentially exposed patient data and disrupted test-result access. Quest paused other systems as a precaution, was still determining what data was accessed, and said providers could use paper requisitions while service was restored.

Japanese gaming company Capcom disclosed a data breach involving nearly 11.8 million records associated with its Capcom ID service. The exposed information includes email addresses, account identifiers, and login activity. Attackers changed passwords on 196 accounts, but Capcom reset those passwords and said it had no evidence that customers' original passwords were compromised. Capcom also said credit-card information was unaffected, contained the unauthorized access, and notified Japanese authorities. The company warned customers to watch for phishing messages that impersonate Capcom, particularly because exposed account and contact information could make those lures more convincing. The briefing did not identify the attacker or explain the initial access method.

Microsoft reported that the China-linked group Storm0940 used an open-source AI penetration-testing framework called Artex to automate attacks against hundreds of organizations. In an October campaign targeting ConnectWise customers, the tool accelerated reconnaissance, password spraying, and abuse of OAuth applications. Microsoft recorded more than 151,000 failed login attempts and successful authentication with compromised ConnectWise application identities. The attackers attempted to reach more than 2,200 Microsoft 365 tenants; in some cases they modified credentials, changed directory settings, and established persistence. One burst targeted more than 300 accounts across eight organizations at over 2,900 attempts per minute. Microsoft said the activity shows how AI can compress the time defenders have to detect established attack methods. Huntress separately described a 10-week Microsoft 365 campaign that compromised accounts at 79 organizations by stealing and replaying authentication tokens. Adversary-in-the-middle phishing captured legitimate sessions, letting attackers bypass MFA without knowing victims' passwords. Huntress saw just under 500 successful sign-ins alongside more than 200,000 failures. Attackers registered 48 rogue devices, replaced authentication methods, and created hidden inbox rules; one account sent a payroll-related message to an outside contact. Huntress recommends conditional access policies and emphasizes that password resets alone are insufficient: organizations must revoke sessions and remove unauthorized devices and authentication methods.

Host Dave Bittner asked Anudeep Parhar, COO of Entrust, whether organizations can rely on an AI “kill switch” as they do with a phone or other device: turn it off if it goes rogue. Parhar said the discussion is useful because autonomous agents can move beyond answering questions to taking actions with real business consequences. But he called a kill switch “the last line in the fence,” not an operating model. If an organization is reaching for it, it may already be too late. Parhar recommended starting with an identity strategy for agents. Organizations should decide whether an agent acts “on behalf of” a named person—such as Dave's agents doing what Dave can do—or belongs to a functional or general-purpose owner. Identity should be followed by authority and authorization: security teams need to connect an action to the human's instruction and intent, verify that the agent is authorized on the target system, and use runtime monitoring to stop a specific transaction through centralized policy rather than shutting down everything. He said recent incidents are often more about human or cloud hygiene than agent autonomy itself. His controls include named ownership, zero trust, least privilege, explicit decision boundaries, and continuous oversight. His advice to boards is to acknowledge that governance is needed while making clear that upstream controls—not a kill switch—are the primary defense.

The episode closes with eMudra's decision to leave the global SSL/TLS certificate business after a security incident drew substantial scrutiny. The company said it revoked the affected certificate authorities within 14 hours and reported no infrastructure compromise, data breach, or customer certificates that had been issued being affected. According to eMudra's chairman, social-media attention was disproportionate because certificates represented only about 1% of the company's revenue. Rising compliance costs and competition from free certificates also contributed to the board's decision to pursue other opportunities. eMudra promised a smooth transition and free replacement certificates for affected customers, while saying its other digital-trust services would continue. The host framed the decision jokingly as eMudra deciding that “SSL certificates just aren't its thing anymore.”