Attackers stole a long-lived npm access token belonging to the lead maintainer of axios, the most popular HTTP client library in JavaScript, and used it to publish two poisoned versions that install a cross-platform remote access trojan. The malicious releases target macOS, Windows, and Linux. They were live on the npm registry for roughly three hours before removal.

Axios gets more than 100 million downloads per week. Wiz reports it sits in approximately 80% of cloud and code environments, touching everything from React front-ends to CI/CD pipelines to serverless functions. Huntress detected the first infections 89 seconds after the malicious package went live and confirmed at least 135 compromised systems among its customers during the exposure window.

This is the third major npm supply chain compromise in seven months. Every one exploited maintainer credentials. This time, the target had adopted every defense the security community recommended.

One credential, two branches, 39 minutes

The attacker took over the npm account of @jasonsaayman, a lead axios maintainer, changed the account email to an anonymous ProtonMail address, and published the poisoned packages through npm’s command-line interface. That bypassed the project’s GitHub Actions CI/CD pipeline entirely.

The attacker never touched the Axios source code. Instead, both release branches received a single new dependency: plain-crypto-js@4.2.1. No part of the codebase imports it. The package exists solely to run a postinstall script that drops a cross-platform RAT onto the developer’s machine.

The staging was precise. Eighteen hours before the axios releases, the attacker published a clean version of plain-crypto-js under a separate npm account to build publishing history and dodge new-package scanner alerts. Then came the weaponized 4.2.1. Both release branches hit within 39 minutes. Three platform-specific payloads were pre-built. The malware erases itself after execution and swaps in a clean package.json to frustrate forensic inspection.

StepSecurity, which identified the compromise alongside Socket, called it among the most operationally sophisticated supply chain attacks ever documented against a top-10 npm package.

The defense that existed on paper

Axios did the right things. Legitimate 1.x releases shipped through GitHub Actions using npm‘s OIDC Trusted Publisher mechanism, which cryptographically ties every publish to a verified CI/CD workflow. The project carried SLSA provenance attestations. By every modern measure, the security stack looked solid.

None of it mattered. Huntress dug into the publish workflow and found the gap. The project still passed NPM_TOKEN as an environment variable right alongside the OIDC credentials. When both are present, npm defaults to the token. The long-lived classic token was the real authentication method for every publish, regardless of how OIDC was configured. The attacker never had to defeat OIDC. They walked around it. A legacy token sat there as a parallel auth path, and npm‘s own hierarchy silently preferred it.

“From my experience at AWS, it’s very common for old auth mechanisms to linger,” said Merritt Baer, CSO at Enkrypt AI and former Deputy CISO at AWS, in an exclusive interview with VentureBeat. “Modern controls get deployed, but if legacy tokens or keys aren’t retired, the system quietly favors them. Just like we saw with SolarWinds, where legacy scripts bypassed newer monitoring.”

The maintainer posted on GitHub after discovering the compromise: “I’m trying to get support to understand how this even happened. I have 2FA / MFA on practically everything I interact with.”

Endor Labs documented the forensic difference. Legitimate axios@1.14.0 showed OIDC provenance, a trusted publisher record, and a gitHead linking to a specific commit. Malicious axios@1.14.1 had none. Any tool checking provenance would have flagged the gap instantly. But provenance verification is opt-in. No registry gate rejected the package.

Three attacks, seven months, same root cause

Three npm supply chain compromises in seven months. Every one started with a stolen maintainer credential.

The Shai-Hulud worm hit in September 2025. A single phished maintainer account gave attackers a foothold that self-replicated across more than 500 packages, harvesting npm tokens, cloud credentials, and GitHub secrets as it spread. CISA issued an advisory. GitHub overhauled npm’s entire authentication model in response.

Then in January 2026, Koi Security’s PackageGate research dropped six zero-day vulnerabilities across npm, pnpm, vlt, and Bun that punched through the very defenses the ecosystem adopted after Shai-Hulud. Lockfile integrity and script-blocking both failed under specific conditions. Three of the four package managers patched within weeks. npm closed the report.

Now axios. A stolen long-lived token published a RAT through both release branches despite OIDC, SLSA, and every post-Shai-Hulud hardening measure in place.

npm shipped real reforms after Shai-Hulud. Creation of new classic tokens got deprecated, though pre-existing ones survived until a hard revocation deadline. FIDO 2FA became mandatory, granular access tokens were capped at seven days for publishing, and trusted publishing via OIDC gave projects a cryptographic alternative to stored credentials. Taken together, those changes hardened everything downstream of the maintainer account. What they didn’t change was the account itself. The credential remained the single point of failure.

“Credential compromise is the recurring theme across npm breaches,” Baer said. “This isn’t just a weak password problem. It’s structural. Without ephemeral credentials, enforced MFA, or isolated build and signing environments, maintainer access remains the weak link.”

What npm shipped vs. what this attack walked past

What to do now at your enterprise

SOC leaders whose organizations run Node.js should treat this as an active incident until they confirm clean systems. The three-hour exposure window fell during peak development hours across Asia-Pacific time zones, and any CI/CD pipeline that ran npm install overnight could have pulled the compromised version automatically.

“The first priority is impact assessment: which builds and downstream consumers ingested the compromised package?” Baer said. “Then containment, patching, and finally, transparent reporting to leadership. What happened, what’s exposed, and what controls will prevent a repeat. Lessons from log4j and event-stream show speed and clarity matter as much as the fix itself.”

• Check exposure. Search lockfiles and CI logs for axios@1.14.1, axios@0.30.4, or plain-crypto-js. Pin to axios@1.14.0 or axios@0.30.3.

• Assume compromise if hit. Rebuild affected machines from a known-good state. Rotate every accessible credential: npm tokens, AWS keys, SSH keys, cloud credentials, CI/CD secrets, .env values.

• Block the C2. Add sfrclak.com and 142.11.206.73 to DNS blocklists and firewall rules.

• Check for RAT artifacts. /Library/Caches/com.apple.act.mond on macOS. %PROGRAMDATA%\wt.exe on Windows. /tmp/ld.py on Linux. If found, preform a full rebuild.

• Harden going forward. Enforce npm ci --ignore-scripts in CI/CD. Require lockfile-only installs. Reject packages missing provenance from projects that previously had it. Audit whether legacy tokens coexist with OIDC in your own publishing workflows.

The credential gap nobody closed

Three attacks in seven months. Each different in execution, identical in root cause. npm’s security model still treats individual maintainer accounts as the ultimate trust anchor. Those accounts remain vulnerable to credential hijacking, no matter how many layers get added downstream.

“AI spots risky packages, audits legacy auth, and speeds SOC response,” Baer said. “But humans still control maintainer credentials. We mitigate risk. We don’t eliminate it.”

Mandatory provenance attestation, where manual CLI publishing is disabled entirely, would have caught this attack before it reached the registry. So would mandatory multi-party signing, where no single maintainer can push a release alone. Neither is enforced today. npm has signaled that disabling tokens by default when trusted publishing is enabled is on the roadmap. Until it ships, every project running OIDC alongside a legacy token has the same blind spot axios had.

The axios maintainer did what the community asked. A legacy token nobody realized was still active and undermined all of it.

CrowdStrike CEO George Kurtz highlighted in his RSA Conference 2026 keynote that the fastest recorded adversary breakout time has dropped to 27 seconds. The average is now 29 minutes, down from 48 minutes in 2024. That is how much time defenders have before a threat spreads. Now CrowdStrike sensors detect more than 1,800 distinct AI applications running on enterprise endpoints, representing nearly 160 million unique application instances. Every one generates detection events, identity events, and data access logs flowing into SIEM systems architected for human-speed workflows.

Cisco found that 85% of surveyed enterprise customers have AI agent pilots underway. Only 5% moved agents into production, according to Cisco President and Chief Product Officer Jeetu Patel in his RSAC blog post. That 80-point gap exists because security teams cannot answer the basic questions agents force. Which agents are running, what are they authorized to do, and who is accountable when one goes wrong.

“The number one threat is security complexity. But we’re running towards that direction in AI as well,” Etay Maor, VP of Threat Intelligence at Cato Networks, told VentureBeat at RSAC 2026. Maor has attended the conference for 16 consecutive years. “We’re going with multiple point solutions for AI. And now you’re creating the next wave of security complexity.”

Agents look identical to humans in your logs

In most default logging configurations, agent-initiated activity looks identical to human-initiated activity in security logs. “It looks indistinguishable if an agent runs Louis’s web browser versus if Louis runs his browser,” Elia Zaitsev, CTO of CrowdStrike, told VentureBeat in an exclusive interview at RSAC 2026. Distinguishing the two requires walking the process tree. “I can actually walk up that process tree and say, this Chrome process was launched by Louis from the desktop. This Chrome process was launched from Louis’s Claude Cowork or ChatGPT application. Thus, it’s agentically controlled.”

Without that depth of endpoint visibility, a compromised agent executing a sanctioned API call with valid credentials fires zero alerts. The exploit surface is already being tested. During his keynote, Kurtz described ClawHavoc, the first major supply chain attack on an AI agent ecosystem, targeting ClawHub, OpenClaw’s public skills registry. Koi Security’s February audit found 341 malicious skills out of 2,857; a follow-up analysis by Antiy CERT identified 1,184 compromised packages historically across the platform. Kurtz noted ClawHub now hosts 13,000 skills in its registry. The infected skills contained backdoors, reverse shells, and credential harvesters; Kurtz said in his keynote that some erased their own memory after installation and could remain latent before activating. “The frontier AI creators will not secure itself,” Kurtz said. “The frontier labs are following the same playbook. They’re building it. They’re not securing it.”

Two agentic SOC architectures, one shared blind spot

Approach A: AI agents inside the SIEM. Cisco and Splunk announced six specialized AI agents for Splunk Enterprise Security: Detection Builder, Triage, Guided Response, Standard Operating Procedures (SOP), Malware Threat Reversing, and Automation Builder. Malware Threat Reversing is currently available in Splunk Attack Analyzer and Detection Studio is generally available as a unified workspace; the remaining five agents are in alpha or prerelease through June 2026. Exposure Analytics and Federated Search follow the same timeline. Upstream of the SOC, Cisco’s DefenseClaw framework scans OpenClaw skills and MCP servers before deployment, while new Duo IAM capabilities extend zero trust to agents with verified identities and time-bound permissions.

“The biggest impediment to scaled adoption in enterprises for business-critical tasks is establishing a sufficient amount of trust,” Patel told VentureBeat. “Delegating and trusted delegating, the difference between those two, one leads to bankruptcy. The other leads to market dominance.”

Approach B: Upstream pipeline detection. CrowdStrike pushed analytics into the data ingestion pipeline itself, integrating its Onum acquisition natively into Falcon’s ingestion system for real-time analytics, detection, and enrichment before events reach the analyst’s queue. Falcon Next-Gen SIEM now ingests Microsoft Defender for Endpoint telemetry natively, so Defender shops do not need additional sensors. CrowdStrike also introduced federated search across third-party data stores and a Query Translation Agent that converts legacy Splunk queries to accelerate SIEM migration.

Falcon Data Security for the Agentic Enterprise applies cross-domain data loss prevention to data agents’ access at runtime. CrowdStrike’s adversary-informed cloud risk prioritization connects agent activity in cloud workloads to the same detection pipeline. Agentic MDR through Falcon Complete adds machine-speed managed detection for teams that cannot build the capability internally.

“The agentic SOC is all about, how do we keep up?” Zaitsev said. “There’s almost no conceivable way they can do it if they don’t have their own agentic assistance.”

CrowdStrike opened its platform to external AI providers through Charlotte AI AgentWorks, announced at RSAC 2026, letting customers build custom security agents on Falcon using frontier AI models. Launch partners include Accenture, Anthropic, AWS, Deloitte, Kroll, NVIDIA, OpenAI, Salesforce, and Telefónica Tech. IBM validated buyer demand through a collaboration integrating Charlotte AI with its Autonomous Threat Operations Machine for coordinated, machine-speed investigation and containment.

The ecosystem contenders. Palo Alto Networks, in an exclusive pre-RSAC briefing with VentureBeat, outlined Prisma AIRS 3.0, extending its AI security platform to agents with artifact scanning, agent red teaming, and a runtime that catches memory poisoning and excessive permissions. The company introduced an agentic identity provider for agent discovery and credential validation. Once Palo Alto Networks closes its proposed acquisition of Koi, the company adds agentic endpoint security. Cortex delivers agentic security orchestration across its customer base.

Intel announced that CrowdStrike’s Falcon platform is being optimized for Intel-powered AI PCs, leveraging neural processing units and silicon-level telemetry to detect agent behavior on the device. Kurtz framed AIDR, AI Detection and Response, as the next category beyond EDR, tracking agent-speed activity across endpoints, SaaS, cloud, and AI pipelines. He said that “humans are going to have 90 agents that work for them on average” as adoption scales but did not specify a timeline.

The gap no vendor closed

The matrix makes one thing visible that the keynotes did not. No vendor shipped an agent behavioral baseline. Both approaches automate triage and accelerate detection. Based on VentureBeat’s review of announced capabilities, neither defines what normal agent behavior looks like in a given enterprise environment.

Teams running Microsoft Sentinel and Security Copilot represent a third architecture not formally announced as a competing approach at RSAC this week, but CISOs in Microsoft-heavy environments need to test whether Sentinel’s native agent telemetry ingestion and Copilot’s automated triage close the same gaps identified above.

Maor cautioned that the vendor response recycles a pattern he has tracked for 16 years. “I hope we don’t have to go through this whole cycle,” he told VentureBeat. “I hope we learned from the past. It doesn’t really look like it.”

Zaitsev’s advice was blunt. “You already know what to do. You’ve known what to do for five, ten, fifteen years. It’s time to finally go do it.”

Five things to do Monday morning

These steps apply regardless of your SOC platform. None requires ripping and replacing current tools. Start with visibility, then layer in controls as agent volume grows.

1. Inventory every agent on your endpoints. CrowdStrike detects 1,800 AI applications across enterprise devices. Cisco’s Duo Identity Intelligence discovers agentic identities. Palo Alto Networks’ agentic IDP catalogs agents and maps them to human owners. If you run a different platform, start with an EDR query for known agent directories and binaries. You cannot set policy for agents you do not know exist.

2. Determine whether your SOC stack can differentiate agent from human activity. CrowdStrike’s Falcon sensor and AIDR do this through process tree lineage. Palo Alto Networks’ agent runtime catches memory poisoning at execution. If your tools cannot make this distinction, your triage rules are applying the wrong behavioral models.

3. Match the architectural approach to your current SIEM. Splunk shops gain agent capabilities through Approach A. Teams evaluating migration get pipeline detection with Splunk query translation and native Defender ingestion through Approach B. Palo Alto Networks’ Cortex delivers a third option. Teams on Microsoft Sentinel, Google Chronicle, Elastic, or other platforms should evaluate whether their SIEM can ingest agent-specific telemetry at this volume.

4. Build an agent behavioral baseline before your next board meeting. No vendor ships one. Define what your agents are authorized to do: which APIs, which data stores, which actions, at which times. Create detection rules for anything outside that scope.

5. Pressure-test your agent supply chain. Cisco’s DefenseClaw and Explorer Edition scan and red-team agents before deployment. CrowdStrike’s runtime detection catches compromised agents post-deployment. Both layers are necessary. Kurtz said in his keynote that ClawHavoc compromised over a thousand ClawHub skills with malware that erased its own memory after installation. If your playbook does not account for an authorized agent executing unauthorized actions at machine speed, rewrite it.

The SOC was built to protect humans using machines. It now protects machines using machines. The response window shrank from 48 minutes to 27 seconds. Any agent generating an alert is now a suspect, not just a sensor. The decisions security leaders make in the next 90 days will determine whether their SOC operates in this new reality or gets buried under it.