Wednesday, November 11, 2026 · 1 MIN READ
Patch Pressure Meets AI Access Risks
IT BRIEFING FOR BUSINESSES
Today’s briefing puts patching and access control at the top of the business IT agenda. Microsoft’s updates include an exploited Outlook vulnerability, while CISA has added six actively exploited flaws to its catalog. Prioritize affected Windows and Office devices, email security systems, and exposed business applications rather than treating every update as equally urgent. For teams managing Microsoft 365 and desktop Office, confirm which installed products need updates and verify deployment. Review internet-facing services and network access, and include backup infrastructure in the review given the reported staging of Veeam exploits.
The AI stories highlight several distinct risks: publicly accessible apps containing sensitive information, stolen provider keys generating unauthorized charges, and agents operating with excessive or poorly separated access. Ask staff to identify AI-built apps and check whether internal data is reachable without authentication. Inventory stored API keys, replace exposed credentials, and review usage charges. Give agents dedicated identities, narrowly scoped permissions, and separate activity records. Research on model backdoors also reinforces the need to control who can modify the models behind agents, especially when those agents can execute commands.
Microsoft Updates Address Exploited Outlook Flaw
Microsoft released a large batch of security updates covering Windows, Office, and other supported software, including an exploited Outlook vulnerability that may expose NTLMv2 hashes. The Hacker News reported 1,259 vulnerabilities, while Krebs on Security reported nearly 1,100 and identified 63 critical flaws. Krebs also noted security updates from Adobe, Apple, Google, and Mozilla.
SOURCES:The Hacker News Krebs on Security
Exploited Flaws Listed as Qlik Servers Face Attacks
CISA added six vulnerabilities to its Known Exploited Vulnerabilities catalog, with the supplied notice naming flaws in Cisco BroadWorks, Barracuda products, and Openwave Webtop Media Transcoder. Separately, Cynet researchers observed unsuccessful attempts to deliver PowerShell reverse shells to Qlik Sense servers across multiple environments; that report also described staging of ActiveMQ and Veeam exploits.
SOURCES:CISA Advisories The Hacker News
Thousands of AI-Built Apps Accessible Without Sign-In
Claroc found 37,571 Grok-built apps online, of which 35,578 were accessible without signing in. Although most were games, landing pages, or calculators, some exposed confidential information and passwords. The publicly accessible apps included 24 created by staff at X and SpaceXSI.
SOURCES:The Hacker News
AI Agent Tests Reveal Identity and Model Risks
Token described building an agent that defaulted to violating its own principles for dedicated identities, task-appropriate access, and activity tracking separate from human users. In separate research, Microsoft demonstrated that modifying fewer than 10 rarely used model neurons could implant a backdoor without retraining; a trigger could then produce an attacker command for an agent to execute with its existing permissions. The technique requires access to modify model weights.
SOURCES:The Hacker News The Hacker News
Stolen AI Keys Linked to $740,000 in Charges
Railway said AI provider keys stolen from customer storage during an October breach were used to generate an estimated $740,000 in unauthorized charges. The application hosting platform said it would cover those charges when the AI providers declined to refund them.
SOURCES:The Hacker News
eMudhra Revokes Eight Additional Certificate Authorities
eMudhra revoked eight more intermediate certificate authorities that it had previously described as unaffected by an October key compromise. The additional revocations followed an incident involving an internal Python script, although the report said the script had not weakened the keys of these eight authorities.
SOURCES:The Hacker News