Alex: Welcome to Cleartext. It's Wednesday, October 21st, 2026. I'm Alex Chen.

Jordan: And I'm Jordan Reeves. Let's get into it.

Alex: We have a packed show today. PHP zero-days being exploited at scale across the internet, a major supply chain compromise through ConnectWise hitting hundreds of Microsoft 365 tenants, Russian GRU running a fresh espionage campaign against NATO's defense industrial base, and the U.S. government formally banning DeepSeek from federal systems. We're also going to talk about a certificate authority that left its private keys so weak anyone could recover them in under a second, and new federal guidance on what to do when your AI agents go rogue. Let's start with the fires that are burning right now.

Jordan: Yeah, let's start with PHP2Shell because this is the one where I want every CISO listening to pause their morning coffee and go check something. Two critical RCE vulnerabilities, CVE-2026-109618 and 109619, CVSS 9.2, affecting the PHP runtime itself. Not your application code. Not your framework. The runtime. If you run PHP on an internet-facing server, you are exposed regardless of how well your application is written.

Alex: And the ugly part of this story is how we got here. These flaws were leaked before the coordinated disclosure window closed. Attackers had exploit details before patches were available. Active in-the-wild exploitation is confirmed.

Jordan: Right. The PHP team has since released emergency patches across four branches: 8.5.12, 8.4.27, 8.3.36, and 8.2.35. There is no workaround. You patch or you're vulnerable. Full stop. If you're running PHP in production, which statistically a huge number of enterprises are somewhere in their stack, this is a drop-everything item today.

Alex: And I want to connect this to the second vulnerability story because they share a common theme, which is that patching delays are being punished faster than ever. The Palo Alto Networks GlobalProtect flaw, CVE-2026-0257, was patched back in May. It's an authentication bypass that lets remote unauthenticated attackers establish VPN connections by bypassing cookie validation, effectively defeating your MFA at the perimeter.

Jordan: Both Qilin and INC ransomware groups have been exploiting this since at least June. That's four months of active exploitation post-patch. Victims span healthcare, education, government, energy. If your organization runs GlobalProtect and you didn't apply the May patch promptly, this isn't a hypothetical. You should be running incident verification right now, not just patching. Look for unauthorized VPN sessions, anomalous authentication patterns, lateral movement indicators. The patch closes the door, but you need to confirm no one already walked through it.

Alex: The broader point for CISOs here is that the window between patch release and weaponization has collapsed to the point where your patch SLA is effectively your breach SLA. If your critical vulnerability patching cycle is still measured in weeks, you're operating on borrowed time.

Jordan: Let's pivot to supply chain, because the ConnectWise story is one that should make every CISO who relies on MSP or MSSP tooling deeply uncomfortable. Attackers compromised an API secret in ConnectWise's cloud backup product and used it to move laterally into hundreds of customers' Microsoft 365 environments. No additional authentication required.

Alex: This is the nightmare scenario with managed service provider tooling. You have a single credential with broad access across your entire customer base, and when that credential is compromised, the blast radius is enormous. Email, files, identity data, all exposed across every affected tenant.

Jordan: The architectural problem here is fundamental. MSP platforms by design hold privileged access to customer environments. That's how they deliver value. But it also means a single compromised API secret becomes a skeleton key. CISOs need to be asking hard questions about API secret governance with their MSP and MSSP partners. How are secrets stored? How are they rotated? What monitoring exists for anomalous API usage patterns? And critically, can you detect if your MSP's credentials are being used in ways that don't match normal operational patterns?

Alex: If you're a ConnectWise customer specifically, you need to be auditing your M365 tenant for unauthorized access right now. Check sign-in logs, review mail forwarding rules, look for consent grants you didn't authorize.

Jordan: Now let's talk about the GRU, because APT28 is back in the headlines and the targeting is pointed. Symantec attributed a campaign hitting a NATO shipbuilder, an aircraft engine manufacturer, and a law enforcement agency. Classic defense industrial base and government targets.

Alex: For CISOs in the defense sector, aerospace, or law enforcement adjacent organizations, this should trigger a direct threat model update. APT28 is one of the most capable and persistent nation-state actors in the world. They have deep resources, sophisticated tooling, and strategic patience. If your organization touches NATO supply chains in any capacity, you are in the target set.

Jordan: What's notable here is the breadth of the targeting. Shipbuilding, aircraft engines, law enforcement. This isn't narrow collection against a single weapons program. This looks like broad industrial espionage designed to map capabilities and supply chains across the NATO defense ecosystem. Symantec has published indicators of compromise and they're urging immediate network scans. Take them up on it.

Alex: And staying in the geopolitical lane, there's an AI-offensive story that deserves attention. A Chinese-speaking threat actor weaponized Cairn, which is a legitimate AI-powered penetration testing tool, to automate attacks that compromised hundreds of websites globally. Victims include John Deere, SharkNinja, and Cummins.

Jordan: This is the story I've been warning about for two years. AI-powered offensive tooling has crossed from theoretical to operational. The attacker didn't build custom AI capabilities. They took a commercially available pen testing tool and turned it into an automated attack platform. The compression of attacker timelines is real. What used to take a skilled operator days now takes hours with AI augmentation. CISOs need to internalize that the pace of attacks against their perimeter is accelerating, and not because attackers got smarter, but because they got better tools.

Alex: Let's move to SafePay, because CISA and the FBI dropped a joint advisory with UK and German counterparts, and the numbers are staggering. Over 600 critical infrastructure organizations breached since 2024, spanning healthcare, energy, and government.

Jordan: Six hundred organizations. That puts SafePay in the top tier of ransomware operators by volume. And their initial access methods are almost insultingly basic: password guessing, vulnerability exploitation, phishing. This isn't sophisticated zero-day exploitation. This is commodity access at scale. The multinational advisory is significant. When CISA, FBI, NCSC, and BSI jointly issue guidance, they're signaling that this group represents a systemic threat that crosses jurisdictional boundaries. Review the TTPs in the advisory and validate your defenses against each one.

Alex: Now, the eMudhra story is one that's easy to overlook but has potentially far-reaching implications. Eight intermediate certificates from eMudhra, an Indian certificate authority that's in major browser trust stores and has issued over 60 million certificates, were secured with private keys so weak they could be recovered from public certificates in under one second.

Jordan: Under one second, Alex. A researcher found this by querying the Common CA Database and demonstrated full private key recovery. These are intermediate CA certificates, which means anyone who recovered those keys could issue arbitrary certificates that would be trusted by browsers and operating systems worldwide. The chain of trust is only as strong as its weakest link, and this is about as weak as it gets.

Alex: If your organization uses eMudhra-issued certificates anywhere, you need an immediate inventory and revocation verification. And more broadly, this is a reminder that certificate authority trust is assumed, not verified, by most enterprises. When was the last time you audited which CAs your systems actually trust?

Jordan: Let's shift to governance. Two big stories here. First, DHS has formally ordered all federal civilian agencies to remove DeepSeek products from their IT systems and banned new purchases. This extends to third-party products developed using DeepSeek implementations.

Alex: This is significant beyond the federal government. When DHS formally classifies a Chinese AI tool as a national security risk, contractor compliance requirements inevitably follow. If you're in the federal supply chain, start planning your DeepSeek removal now. But even outside government contracting, this sets a regulatory precedent that enterprise CISOs in regulated industries should use to accelerate their own AI vendor risk and data sovereignty policies. If you've been struggling to get executive buy-in to restrict certain AI tools, the DHS order just gave you a powerful reference point.

Jordan: And on the AI governance front, NSA, NIST, and CISA jointly issued guidance on rogue AI agents. It's a five-stage framework covering detection, containment, and recovery from incidents where AI agents behave in unintended ways. Notably, adversarial attacks are explicitly out of scope. This is about misalignment and unintended behavior.

Alex: This is important framing. The federal government is telling organizations that your own AI agents misbehaving is a security event, not just an engineering bug. If you're deploying agentic AI, and increasingly enterprises are, you need incident response playbooks that cover scenarios where your agents act outside their intended boundaries.

Jordan: Which connects neatly to Pluto Security emerging from stealth with forty-two million dollars specifically targeting agentic AI workflow protection. They're using endpoint telemetry and local models to detect agent intent, identify prompt injection, and enforce security policies. The fact that investors are pouring this kind of money into agentic AI security tells you where the market sees the next major attack surface.

Alex: Agreed. CISOs deploying agentic AI should be evaluating this as a distinct security category, not an extension of existing application security.

Jordan: So stepping back and looking at the week's theme, what jumps out to me is the speed gap. The PHP zero-days leaked and exploited before patches existed. The PAN-OS flaw patched in May and exploited since June. AI tools compressing attacker timelines. Across every story today, the through-line is that the tempo of offensive operations is outpacing most organizations' defensive response cycles.

Alex: And the answer isn't just faster patching, though that's necessary. It's architectural. Assume breach, segment aggressively, monitor for post-compromise activity, and don't trust single points of credential authority, whether that's your MSP's API secret or a certificate authority's private key. The organizations that will weather this environment are the ones that have built resilience, not just defenses.

Jordan: Well said. The wall won't hold. Build the response muscle.

Alex: That's our show for Wednesday, October 21st. Show notes and links to every story we covered are at cleartext.fm. We'll see you tomorrow.

Jordan: Stay sharp.