CSO Online reports AI speeds up attacks, core techniques unchanged
CSO Online reports AI is raising the speed and scale of attacks while core techniques stay the same; patch exposed Langflow and speed up patching.
- Published
- Ecosystem
- AI developer toolsOpen sourceCloud
- Exploitation
- Yes, according to CSO Online, which reports that the JadePuffer campaign used CVE-2025-3248 for initial access; no independent verification in the sources
Live scores
CVE-2025-3248- EPSS
- 1.000
very likely to be attacked soon - VEDAS
- 0.983
strong exploit evidence - On CISA KEV
- Yes, since 2025-05-05
Thin horizontal line: the “likely exploited” threshold (EPSS 0.10, VEDAS 0.5). Shaded “no data” areas are periods when cve-scores did not publish. The vertical line is the day CISA added it to its exploited list.
VEDAS flagged this bug about 15 days before CISA confirmed attacks.
What happened
According to a feature by John Leyden in CSO Online, AI is helping attackers increase the speed and scale of their operations, but it has not yet changed their core objectives or the techniques in their attack chains. CSO Online says, citing security experts, that phishing, malware and vulnerability exploitation remain the main techniques, while AI abuse has made attacks easier and cheaper to mount for relatively unskilled attackers. This is a single-source report of expert opinion, and none of the claims below are independently verified.
CSO Online cites two examples:
- GTG-1002: CSO Online describes this as the most high-profile AI-powered attack to date, in which Chinese state-sponsored actors abused Claude Code in an AI-orchestrated cyber-espionage campaign in September 2025. The article says conventional techniques laid the foundation and AI supported reconnaissance, vulnerability discovery, exploitation, lateral movement, credential harvesting, data analysis and exfiltration.
- JadePuffer: CSO Online reports this ransomware campaign was disclosed in July 2026, with an LLM-driven agent automating most of the intrusion chain. According to the article, initial exploitation used CVE-2025-3248, a year-old missing-authentication flaw in Langflow, an open-source framework for building LLM applications and agent workflows. The article says vulnerable internet-facing Langflow instances were looted for cloud credentials, API keys and other secrets, followed by lateral movement, privilege escalation and destructive database encryption.
Quoted experts in the article make these claims:
- Ronald Lewis of Black Duck says the attack stages are unchanged, but AI changes “the volume and velocity” of the process.
- Crystal Morin of Sysdig says AI-assisted attackers are showing signs of shortening detection windows, and that agentic attacks are still flawed. She says JadePuffer had flaws in its ransom note, encryption, wallet and data exfiltration.
- Nazar Tymoshyk of UnderDefense uses the term “agentic APT” and says it is no longer exclusive to nation-state actors. He also says attackers favor Chinese models such as GLM and Kimi 3. Darren Meyer of Checkmarx disputes this kind of model attribution, calling it “pure speculation from very thin evidence on the ground.”
- Nathan Smolenski of Qualys says the window between disclosure and exploitation is shrinking to near zero, and that 30-day patch cadences can no longer be relied on.
Not confirmed in the source: the fixed Langflow version, the number of affected instances, the number of victims, and any technical indicators for JadePuffer.
Am I affected?
The article does not give a product or version list, so exposure is not confirmed for any specific organization. According to CSO Online, the clearest concrete exposure is internet-facing Langflow instances that were vulnerable to CVE-2025-3248. If you do not run Langflow, the article’s points are general trend observations rather than a specific vulnerability in your stack.
What to do now
- Inventory any Langflow deployments, especially internet-facing ones, and confirm their patch status for CVE-2025-3248. The fixed version is not confirmed in the source, so check the project’s own advisory.
- If an exposed instance may have been unpatched, consider rotating the cloud credentials, API keys and other secrets it could access. CSO Online reports these were looted in JadePuffer.
- Review your patch cadence. Smolenski of Qualys, quoted by CSO Online, says human-speed 30-day cycles can no longer be relied on.
- Keep foundational controls such as patching and authentication in place. CSO Online’s experts also suggest automating lower-level triage with human-in-the-loop guardrails.