LIVE
SECURITY

Global agencies battle AI-discovered PHP2Shell web flaws as Microsoft warns 10 neurons can backdoor AI agents

Emergency patching continues after autonomous AI scanners found critical vulnerabilities in PHP—used by 70% of websites—while Microsoft Research demonstrates the Lazarus agent backdoor.

UpdateNext Staff
Editorial team
3 min read
1 view
Cybersecurity agencies respond to AI-discovered PHP2Shell vulnerabilities and Microsoft Lazarus agent backdoor research — UpdateNext
Automated AI vulnerability discovery triggered global PHP patching alerts and temporary government portal shutdowns. Photo: CTV News

What is the PHP2Shell vulnerability wave?

System administrators worldwide are racing to complete emergency patch cycles following the discovery of PHP2Shell—a set of deep remote-code-execution flaws in the PHP runtime uncovered by autonomous AI bug-hunting models. Because PHP underpins roughly 70% of active websites and content management systems, CISA and national cyber authorities issued urgent remediation directives, and several municipal and national web portals in Europe and Asia briefly went offline to apply fixes.

ADVERTISEMENT
Get the daily AI briefingSubscribe

How does Microsoft’s 10-neuron Lazarus backdoor work?

Adding to concerns over agent security, Microsoft Research published a new study detailing the "Lazarus" technique, demonstrating that an attacker who modifies fewer than 10 dormant neurons inside an open-weight language model can implant a sleeper backdoor. The compromised agent behaves normally on all standard safety benchmarks until triggered by a specific input phrase.

✓ WE KNOW
  • AI-driven vulnerability scanners uncovered the PHP2Shell flaws affecting PHP web servers globally.
  • Microsoft Research demonstrated that modifying fewer than 10 dormant neurons can backdoor an LLM agent.
? STILL UNCLEAR
  • How many unpatched legacy PHP servers remain exposed to automated exploitation.
SOURCES:CTV News: Websites shut down globally after AI finds cybersecurity flaw affecting 70% of the internet ↗The Guardian: AI hacking tools and the PHP vulnerability tsunami ↗AI Weekly: Microsoft Research Lazarus 10-Neuron Agent Backdoor Study ↗
> how_we_reported_this
  • ✓Sources: written from primary sources and official statements, linked above.
  • ✓Fact-check: names, dates and claims checked against the original sources.
  • ✓AI use: researched with AI tools; written and reviewed by our editors. Our AI policy
  • ✓Independence: UpdateNext has no financial relationship with the companies covered.
1 view
> react:
UpdateNext Staff
The UpdateNext newsroom covers AI and tech from primary sources, every day. Every story is checked against official statements.
Write for us →
Comments
Comments are moderated · Sign in to comment
> read_next
PATCH TUESDAY ZERO-DAY
Microsoft addresses 1,000+ vulnerabilities and patches actively exploited Outlook NTLMv2 zero-day (CVE-2026-124194) found with Claude AI
WINDOWS ZERO-DAY ALERT
Public 'SignDrop' zero-day exploit targets Windows CLFS driver (clfs.sys) companion files to seize SYSTEM privileges
CISA KEV MANDATE
CISA adds Check Point Gaia OS DynamicCLI root zero-day (CVE-2026-123039) to KEV catalog as attackers deploy ESXi wipers
What's next in AI.

The daily briefing on AI news, tools and breakthroughs.

Agencies Battle AI-Found PHP2Shell Flaw & 10-Neuron Backdoors