In partnership with
☀️ Good morning. Here's everything that happened in cybersecurity yesterday — in under 5 minutes.
Artificial intelligence sat on both sides of the ledger yesterday: Palo Alto's Unit 42 said its NOVA system autonomously surfaced more than 14,000 unknown vulnerabilities across the open-source supply chain, while Google was forced to delete three AI agent workflows after Pillar Security demonstrated the first documented agent-on-agent attack. But the bugs actually being exploited were decidedly human-made — CISA added an already-exploited N-able N-central authentication bypass (CVE-2026-18577) to its Known Exploited Vulnerabilities catalog, cPanel patched a critical CVSS 9.4 flaw that let a hosting customer run SQL as database root, and our monthly CVE Watch closed the book on a July that shipped a record 569-CVE Patch Tuesday and four SharePoint zero-days. The signal underneath it all: the AI headlines are loud, but the KEV list is where your morning should start.
🔥 Top Stories
01 — Unit 42's NOVA System Surfaces 14,000+ AI-Discovered Zero-Days Across the Open-Source Supply Chain
Artificial Intelligence
Palo Alto Networks' Unit 42 disclosed that its NOVA system — an autonomous, frontier-AI vulnerability-discovery pipeline — surfaced more than 14,000 previously unknown flaws across open-source software, a scale the firm frames as the leading edge of AI-industrialized zero-day discovery. The findings span the dependency graph that virtually every modern application inherits, and Unit 42 published the work under the banner of a "frontier AI vulnerability burst."
The hard question isn't whether the flaws are real — it's who triages them. Fourteen thousand machine-generated findings land on maintainers who are, in most projects, unpaid volunteers, and the same week saw warnings that AI is already polluting the CVE pipeline with fake reports. Treat this as a preview of the coming remediation bottleneck: the constraint is shifting from finding bugs to validating and fixing them, so weight your program toward triage capacity and provenance, not just scanner coverage.
02 — Google Deletes 3 ADK AI Workflows After Pillar Security Discloses an Agent-to-Agent Attack
Artificial Intelligence
Google removed three workflows from its Agent Development Kit (ADK) Python repository after Pillar Security showed that a poisoned public GitHub issue could manipulate a low-privilege triage agent into triggering a privileged code-fixing agent. Because the triage agent was a recognized collaborator, its planted /adk-issue-fix comment satisfied the trust check — handing the attacker a path that exposed secrets and enabled pull-request tampering.
This is being called the first documented agent-on-agent attack, and it reframes prompt injection as a privilege-escalation problem rather than a chatbot curiosity. Any pipeline where one agent can hand off to a more-privileged agent needs the same scrutiny you'd give a service account: scope permissions tightly, treat inter-agent hand-offs as untrusted input, and never let a public issue or PR comment become an execution trigger.
03 — CISA Adds the Exploited N-able N-central Flaw (CVE-2026-18577) to Its KEV Catalog
Vulnerabilities
CISA added CVE-2026-18577 (CVSS 8.2) to its Known Exploited Vulnerabilities catalog after N-able found a second authentication-bypass vector past the incomplete fix for CVE-2026-18556. The flaw lets a remote, unauthenticated attacker seize administrative control of vulnerable N-central servers — a Remote Monitoring and Management platform sitting at the center of countless MSP environments.
An RMM compromise is a supply-chain compromise: administrative control of N-central can cascade to every downstream customer the provider manages. KEV listing puts federal agencies on a binding remediation clock, but every MSP running N-central is in scope. Patch to the fixed build now, hunt for signs of prior exploitation, and rotate credentials on any server that was internet-exposed before the update.
04 — cPanel Patches CVE-2026-58048 (CVSS 9.4): A Hosting Customer Could Run SQL as Database Root
Vulnerabilities
cPanel shipped a targeted security release for CVE-2026-58048 (CVSS 4.0 score 9.4), where an authenticated hosting customer could execute SQL in the database's root context — crossing the boundary between an ordinary cPanel account and the server's administrative database identity. The same release closed two additional routes past account boundaries.
On shared hosting, that privilege boundary is the whole security model: one tenant reaching database-root can read or alter every other tenant's data on the box. There's no meaningful workaround, so the action is simply to update to the patched cPanel version immediately — hosting providers first, since a single unpatched server exposes all of its customers at once.
05 — The Top CVEs of July 2026: Four SharePoint Zero-Days and a Record 569-CVE Patch Tuesday
CVE Watch
July belonged to SharePoint: four separate flaws were exploited in the wild against on-premises deployments, while Microsoft shipped the largest Patch Tuesday in its history at 569 CVEs. Our monthly CVE Watch ranks what mattered and, more usefully, what to fix first when the list is that long.
The volume is the story — 569 fixes in a single cycle is a triage problem no team can brute-force. Anchor on exploited-in-the-wild and KEV-listed items (SharePoint chief among them), then work down by exposure and CVSS. The Swiss federal IT agency breach reported yesterday, still under investigation with SharePoint flaws suspected, is a live reminder that these on-prem bugs are being turned into intrusions right now.
⭐️ Our Partner
Own Search With Podcasts
Your competitors are fighting over the same keywords. The smartest brands are building the authority that search engines, AI platforms, and customers trust everywhere.
Every relevant podcast appearance can produce branded mentions, backlinks, transcripts, citations, clips, expert content, and third-party proof that keeps compounding across search and AI discovery.
PodPitch searches millions of podcasts, finds the shows that matter to your market, develops the angle, sends personalized pitches, and follows up automatically until your experts are booked.
Growth teams are already using podcast appearances to build distributed authority that cannot be manufactured by publishing another generic SEO article.
Only 20 SEO, AEO, and GEO demo spots are available this month. Once they’re claimed, the offer disappears.
Start building searchable authority now, before your competitors own the conversations shaping your market.
📊 By The Numbers
14,000+ — Previously unknown vulnerabilities surfaced across the open-source supply chain by Unit 42's NOVA AI system.
569 — CVEs in Microsoft's July Patch Tuesday, the largest single-month release in company history.
CVSS 9.4 — Severity of cPanel's CVE-2026-58048, letting a hosting customer execute SQL as database root.
CVSS 8.2 — Score of N-able N-central's CVE-2026-18577, now on CISA's KEV list after active exploitation.
3 — AI agent workflows Google deleted from its ADK repo after the agent-to-agent attack disclosure.
⚡ The Signal
Yesterday reads like a referendum on where AI actually lives in security work — and the answer is "everywhere except the part that fixes things." NOVA can autonomously generate 14,000 findings; ADK agents can be turned against each other; and vendors spent the day announcing AI pentesting, AI red-teaming, and autonomous remediation. Offense and discovery are scaling almost frictionlessly.
Remediation is not. The two vulnerabilities that were actually being exploited yesterday — the N-able auth bypass and, by strong suspicion, SharePoint at the Swiss federal IT agency — are ordinary, human-authored bugs with human-authored patches. AI didn't find them and AI isn't fixing them; people are, on a clock, one server at a time. cPanel's database-root flaw is the same shape: a boundary error a human wrote and a human must patch.
That's the through-line worth carrying into your week. The industry's AI narrative inflates the top of the funnel — more findings, more clever attacks, more tooling — while the bottleneck stays stubbornly at the bottom, where a technician still has to apply the update. When AI-generated findings and AI-fabricated CVEs both flood in, the scarce resource becomes trustworthy triage: knowing which of 14,000 flaws is real, which of 569 is being exploited, and which server to touch first. Start with the KEV list, not the press release.
🔍 What You May Have Missed
What Is Privileged Access Management (PAM)? — A ground-up guide to why privileged accounts warrant special treatment and how to stand up a PAM program — the exact control that would blunt yesterday's RMM and database-root escalations.
Russian hackers abuse hotel Wi-Fi to steal Microsoft 365 credentials — Microsoft named the Midnight Blizzard campaign "CaptiveCrunch," using hijacked hotel captive portals to push the CornFlake and ChocoShell malware strains at travelers.
Device code phishing up 1,500% in 2026, vishing doubles — Newer social-engineering techniques are sidestepping entrenched MFA controls while leaving less forensic evidence behind.
📅 What to Watch
Senate to debate a package of privacy, AI and kids-safety bills — A key committee moves to clear major legislation, headlined by the Kids Online Safety Act, that could reshape how minors interact with online platforms.
Swiss federal IT agency breached, SharePoint flaws suspected — Roughly 200 accounts were compromised after anomalies on on-premises Microsoft servers; the investigation into the entry point is ongoing.
Apple mounts a new legal challenge to the UK over iCloud access — Apple is again fighting British demands for a way around Advanced Data Protection, a case with global privacy implications.
TP-Link Omada flaws chain into full network takeover — Forescout disclosed 15 new vulnerabilities in the Omada ecosystem; patch and watch for exploitation of the zero-touch-provisioning chain.
Stay sharp. Stay ahead.
Till next time,
The CyberSignal Team