Black Kite Flags AI Vendors Amid Active Exploits

Chapters
The short version
Esecurityplanet reported that one in three AI suppliers has critical unpatched vulnerabilities, citing Black Kite research. For smaller firms that buy AI tools from outside vendors, The Tech Buzz reported that a flaw in one vendor's AI system can create an entry point to every customer in that vendor's network.
- Railway said thieves used stolen customer AI credentials to run up approximately $740,000 in fraudulent usage.
- Gbhackers relayed a CISA warning about active attacks on a WordPress core flaw, with over 240,000 attempts logged.
- Google Threat Intelligence Group data showed disclosed vulnerabilities climbing from 5,045 in January to 10,477 in July 2026, The Straits Times said.
- Qualys wrote that Mandiant found attackers putting agentic AI to work for reconnaissance and vulnerability validation.
Black Kite finds critical gaps at AI suppliers
Cyber risk intelligence firm Black Kite found that one in three AI suppliers has critical unpatched vulnerabilities, Esecurityplanet reported on November 11, 2026. A weekly roundup that Esecurityplanet published on November 13, 2026, framed the week around active exploits, AI risks and breaches at third parties.
The Tech Buzz, which covered the research on November 11, 2026, described Black Kite as a third-party cyber risk intelligence platform. Enterprise supply chains face new threats as AI adoption outpaces security controls, The Tech Buzz reported. Many organizations deploying AI agents and large language models lack frameworks to assess whether their vendors' AI implementations meet basic security standards, according to the outlet. A vulnerability in one vendor's AI system can create an entry point to every customer in that vendor's network, The Tech Buzz said.
Mandiant puts the time to exploit below zero
Mandiant's M-Trends 2026 report puts the mean time to exploit a vulnerability at negative seven days, The Straits Times reported on November 12, 2026. The Straits Times said the Mandiant report draws on over 500,000 hours of global incident investigations in 2025.
Qualys cited the same Mandiant estimate in a November 13, 2026, post on Blog Qualys written by its Vice President, Product Management, Platform and Technologies. "AI has changed how fast attackers exploit far more than what they exploit," Qualys wrote. Attackers used agentic AI for reconnaissance, vulnerability validation and, in at least one case, a zero-day, according to Mandiant's AI Risk and Resilience 2026 report as summarized by Qualys. For the sixth year running, vulnerability exploitation remained the leading way in for attackers, Mandiant found, Qualys reported.
Disclosed vulnerabilities more than doubled in 2026, rising from 5,045 in January to 10,477 in July before reaching 10,740 in August, according to Google Threat Intelligence Group figures cited by The Straits Times.
Active exploits hit WordPress and Zyxel gear
CISA warned of active attacks exploiting a WordPress core code-execution flaw, with over 240,000 attempts logged, Gbhackers reported on November 10, 2026, in a newsletter covering November 2 to 6, 2026. Hackers exploited a Zyxel GS1900 zero-day to compromise nearly 1,000 network switches, Gbhackers reported. Attackers also used stolen Microsoft Entra ID app credentials to break into 27 organizations and steal data, according to Gbhackers.
The ARTEX AI framework, tied to bank breaches and cloud intrusions, was found exposed across hundreds of servers, Gbhackers said. ARTEX hackers used Claude Code, DeepSeek and Grok in attacks targeting South Korean banks, and a separate Claude AI flaw let malicious browser extensions execute commands on remote computers, the outlet added. The CISA warning, the Entra ID intrusions and the ARTEX activity were reported by Gbhackers; the reports have not been confirmed elsewhere.
Stolen AI credentials drove fraud at Railway
Railway said attackers stole customer AI credentials and generated approximately $740,000 in fraudulent usage, Esecurityplanet reported on November 12, 2026; the report has not been confirmed elsewhere. The company has committed to covering losses that AI providers will not refund, according to Esecurityplanet.
A breach at Fireworks AI exposed Notion AI chat transcripts, Esecurityplanet reported in a separate article on November 11, 2026; the report has not been confirmed elsewhere.
Hackers used ARTEX to steal Azure Key Vault secrets just five minutes after finding exposed GitHub credentials, Gbhackers reported. In the same weekly newsletter, Gbhackers said LLMJacking ran up six-figure cloud bills; neither Gbhackers report has been confirmed elsewhere.
Vendor flaws fed a Singapore breach wave
Data leaks have surfaced continuously in Singapore since late October, hitting organizations including Fullerton Health, telco M1, SPH Media and the Agency for Integrated Care, The Straits Times reported; the report has not been confirmed elsewhere. Global aviation tech vendor Amadeus compromised Singapore Airlines' passenger records, according to The Straits Times, an account also not confirmed by other outlets.
The Straits Times traced the incidents to opportunistic scanning for misconfigurations and flaws, weaknesses at vendors and coordinated threat campaigns. Artificial intelligence was a primary contributor because it has sped up the discovery of system vulnerabilities, the newspaper said. Payment card and banking details were uncompromised in nearly all of the recent cases, The Straits Times said, but exposed NRIC numbers, full names, dates of birth and phone numbers give criminals the details for targeted impersonation scams.
Tron's take
My take: for a small or mid-sized business, the Black Kite finding deserves more attention this week than any model launch, because the exposure it describes sits with suppliers, outside a firm's own servers.
I do not read the week as a reason to freeze AI use, and I do not read it as a reason to chase every new release. Some owners argue a business must adopt each release immediately to stay competitive; others argue AI news has nothing to do with a small business. My position is that firms do better applying last quarter's proven tools well while tracking what ships, so adoption is timed on purpose. That is my reading of the news, not a reported result.
My advice is three concrete checks. Ask each AI vendor in writing which critical vulnerabilities in its product are unpatched today and the date each will be fixed. Put spending caps and billing alerts on every AI provider account, and rotate any API key stored in a code repository, given Railway's account of stolen customer credentials and the ARTEX theft that began with exposed GitHub credentials. Confirm every company WordPress site runs a current, patched version.
XL.net sells security assessments and managed IT services, and a vendor review tied to the Black Kite finding is where I would start. Readers who followed earlier coverage of AI key abuse will recognize the credential pattern in Railway's account.
Questions I'd expect
What did Black Kite find about AI suppliers?
Black Kite found one in three AI suppliers has critical unpatched vulnerabilities, Esecurityplanet reported. The Tech Buzz reported that many organizations lack frameworks to check whether vendors' AI implementations meet basic security standards.
How quickly are attackers exploiting vulnerabilities?
Mandiant's M-Trends 2026 report estimates the mean time to exploit a vulnerability at negative seven days, The Straits Times reported. Qualys cited the same Mandiant estimate in its November 13, 2026, post.
What happened at Railway?
Railway said attackers stole customer AI credentials and generated approximately $740,000 in fraudulent usage, and it committed to covering losses AI providers will not refund, Esecurityplanet reported. The report has not been confirmed elsewhere.