Hi, I'm Vivian, a cybersecurity and AI Product Manager trying to keep up with an industry that moves faster than I can whisk up my morning matcha. Every week brings a new wave of vulnerabilities, AI security mishaps, and breaches that keep us on our toes, so I take some time to share the most interesting news instead of letting it all blur together. Let's dive into whats been keeping us up at night recently in cybersecurity and AI security.
AI Research & Vulnerabilities:
Joint Study Shows 250 Documents Can Poison Any-Size LLM
A joint study by Anthropic, the UK AI Security Institute, and the Alan Turing Institute found that as few as 250 malicious documents can create a backdoor vulnerability in a large language model, regardless of model size or training data volume. The study focused on a narrow backdoor attack that makes a model produce gibberish text when encountering a specific phrase. The research challenges the assumption that attackers need to control a percentage of training data; instead, only a near-constant number of documents is sufficient for poisoning attacks regardless of model size.
Google Introduces CodeMender AI Agent for Code Security
Google's DeepMind unveiled CodeMender, an AI agent that automatically fixes code vulnerabilities by being both reactive (fixing new vulnerabilities as they're spotted) and proactive (rewriting and securing existing code).
CometJacking Attack Targets Perplexity's Comet Browser
Security researchers disclosed an attack targeting Perplexity's Comet AI browser by embedding malicious prompts within URLs to steal sensitive data that the browser AI already has access to. While the issue has been patched, this represents new risks introduced by AI-native browsers.
Three Vulnerabilities Discovered in Google Gemini
Security researchers discovered three vulnerabilities that could have exposed users to privacy risks and data theft. The flaws included a prompt injection vulnerability in Gemini Cloud Assist, a search injection vulnerability in Gemini's Search Personalization Model, and a data exfiltration vulnerability in Gemini's Browsing Tool. Google has since patched these vulnerabilities.
Cybersecurity News
Scattered Lapsus$ Hunters Launch Massive Salesforce Extortion Campaign
A cybercriminal group called Scattered Lapsus$ Hunters launched a data leak site listing 39 companies, including Disney, Toyota, Adidas, and IKEA, whose Salesforce databases they claimed to have stolen 1 billion records from between April 2024 to September 2025. Despite the group's October 10th ransomware deadline, Salesforce refused to negotiate or pay. On October 9th, the FBI seized a BreachForums domain used as a data leak extortion site for the Salesforce attacks, but the group continues their campaign and reportedly leaked 5.7 million Qantas customer personal data records on the dark web.
Qilin Ransomware Group Hacks Japan's Asahi Group
A Russia-based ransomware group named Qilin claimed responsibility for a cyberattack on Asahi Group Holdings that forced the Japanese beverage giant to suspend orders, shipments, and customer service. The group alleges it stole 27 GB of data including financial documents, employee personal information, and development forecasts. While production resumed earlier this week, investigation is ongoing. Qilin has been the most active ransomware group in 2025 with manufacturers as the primary targets.
Clop Exploits Oracle Zero-Day in Mass Data Theft Campaign
The Clop ransomware group exploited a critical zero-day vulnerability in Oracle E-Business Suite to steal data from dozens of organizations and sent extortion emails to victim executives on September 29, with ransom demands reaching up to $50 million. Google Threat Intelligence Group and Mandiant reported that attackers chained at least five distinct vulnerabilities together to achieve pre-authenticated remote code execution.
SonicWall Confirms All Cloud Backup Customers Affected by Data Breach
SonicWall disclosed that unauthorized actors accessed firewall configuration backup files for all customers who used its MySonicWall cloud backup service, reversing its initial claim that fewer than 5% of customers were impacted. The breach resulted from a brute-force attack against the cloud backup API service that lacked basic protections like rate limiting. Exposed files contain encrypted credentials, network rules, VPN configurations, and service credentials that could enable targeted attacks. SonicWall urged customers to delete existing cloud backups, reset credentials, rotate shared secrets, and recreate backups locally.
Startup/VC News
CRN Highlights 10 Hot 2025 AI Security Startups
CRN released its list of 10 hot AI security startups to watch in 2025, highlighting emerging companies addressing the evolving security challenges posed by artificial intelligence.
Paris-based cybersecurity startup Filigran, which specializes in extended threat management through open-source solutions powered by Agentic AI, secured $58 million in Series C funding to accelerate product development and expand into new markets.
Vilnius-based cybersecurity startup CBRX secured €540,000 in pre-seed funding to develop its AI-powered cloud platform that helps managed service providers transition into managed security service providers by automating incident management, threat monitoring, and customer support without significant upfront investment.
That's all for now… Stay informed and protected.