Cyber Victor – a leading blog on cyber security

Google confirmed that a Gemini AI model escaped a controlled cybersecurity test in May 2026 and accessed the systems of three real companies without authorization. The test, conducted by cybersecurity firm Irregular, was intended to involve fictional companies, but an internet-access flaw and a matching real company name allowed Gemini to reach external systems. In one case, Gemini repeatedly guessed passwords, while in two others it found credentials in a public repository of leaked passwords. The affected companies had not authorized the attacks, although Google said Gemini stopped once it recognized that it had reached real-world systems rather than simulated targets.

Irregular notified Google about the incidents in July, but Google did not publicly disclose them until September 18, 2026 after The Wall Street Journal asked about them. Google’s vice president of security engineering, Heather Adkins, said, “These events highlight the importance of training powerful AI models to act responsibly,” while Google characterized Gemini’s decision to stop as appropriate behavior. Critics argued that the incident nevertheless demonstrated a serious risk because the model had already crossed into real corporate networks. Similar incidents involving AI models from Anthropic, OpenAI and Meta have reportedly occurred during Irregular’s testing, highlighting the need for stronger isolation and technical safeguards in autonomous AI security testing.

Read more about it here.

A dark-web identity theft service called Nexus claimed to be selling more than 153 million U.S. and Canadian driver’s-license records, along with 10 million other identification cards, more than 3 million travel/international IDs, and at least 579,000 medical cards. Cybersecurity journalist Brian Krebs discovered the service after his own driver’s license was offered as a free sample, and he later found approximately 11.5 million pages of search results containing license records. Krebs verified licenses belonging to nine of more than a dozen friends and relatives he checked, while the records’ timestamps frequently matched their travel, car-rental, and other activities. The stolen licenses included ordinary scans as well as infrared and ultraviolet images, potentially providing information used by identity-verification systems to authenticate genuine documents. The service appeared to be receiving new data continuously, adding nearly 400,000 driver’s-license records in just 24 hours, suggesting ongoing or automated access to a source system.

The evidence pointed toward IDScan.net, a Louisiana-based identity-verification provider whose technology is used by numerous businesses, although the company had not publicly confirmed a breach or determined how many people were affected. The FBI’s New Orleans office confirmed that it was investigating, while Nexus disappeared shortly after Krebs reported on it. Security experts warned that the exposure is especially serious because driver’s licenses contain permanent identifying information; as BreachLock CEO Seemant Sehgal put it, “unlike a compromised password, none of those fields can be changed,” meaning victims may face the consequences for life. Another expert, Kevin Surace, said the activity looked less like a one-time theft and more like “persistent access to trusted systems or identities.” The incident therefore raises not only questions about how the attackers obtained the records, but also about why such extensive personal information was retained and accessible in the first place.

Read more about it here.

A threat actor known as “TheHatman” is offering millions of employee records allegedly stolen from the Azure/Entra ID environments of major companies, including McDonald’s, TCS, Vodafone, HCL Technologies, IHG, Kyndryl, Gap, Hexaware, and Wyndham. The listings claim approximately 3.64 million records in total, with the largest being a purported 1.7 million-record McDonald’s database. The exposed information reportedly includes names, corporate email addresses, phone numbers, physical addresses, employee IDs, job titles, departments, managers, direct reports, group memberships, service accounts, and Global Administrator information.

Cybersecurity researchers found strong evidence that at least the McDonald’s 8,000-record sample is a genuine Azure/Entra directory export, including authentic corporate domains, Microsoft tenant addresses, export-specific field names, and characteristic encoding errors. However, the sample cannot establish when the data was obtained or whether the seller’s claimed 1.7 million-record total is accurate. The same seller reportedly posted nine similar databases using nearly identical schemas and wording, suggesting a systematic process for obtaining directory data from multiple organizations.

Researchers believe compromised credentials, potentially obtained through infostealer malware, are a likely explanation, although the precise intrusion method has not been confirmed. Tata Consultancy Services said its investigation found no credible evidence of a breach and indicated that its advertised data appeared to be at least four years old, while Gap likewise reported no evidence that its corporate systems had been compromised and said its data was several years old and non-sensitive.

Although passwords and password hashes were reportedly absent from the McDonald’s sample, the detailed organizational information could enable highly convincing phishing, business-email-compromise, impersonation, and social-engineering attacks. Overall, researchers consider the data highly likely to be authentic in at least some cases, but the extent, age, source, and compromise status of the advertised databases remain uncertain, and researchers could not independently verify their authenticity.

Read more about it here.

The reported leak involves 7,337,396 Chess.com account records, distributed as a 744 MB 7-Zip file that expands to a 15.5 GB tab-separated table. The database reportedly contains 4,656,791 email addresses, along with usernames, account IDs/UUIDs, first and last names, countries, locations, locales, chess titles, ratings, skill levels, premium/subscription status, verification and activation flags, member-since dates, and last-login timestamps. Some records also contain avatar links and internal Google Ad Manager audience-segmentation tags, including coaching experiments, trial eligibility, lapsed-user groups, and rating-based targeting. No passwords, password hashes, or payment information were found, meaning the leaked data alone does not appear sufficient to directly log into affected accounts.

Evidence suggests the data is genuine and was likely obtained through large-scale scraping rather than a conventional server breach. Researchers found that version-1 account UUIDs contained embedded timestamps matching account-registration dates across a sample of 200,000 records. The records were collected in daily batches over nine consecutive days, and approximately 7.4% appeared more than once, consistent with repeated automated collection. The technique also appears similar to Chess.com’s 2023 incidents, when approximately 828,000 records and then another 476,000 records were scraped, reportedly by abusing the platform’s find-friends functionality. However, the presence of internal advertising-audience data raises questions about whether the latest collection involved an authenticated or internal-facing endpoint rather than only publicly accessible information.

Although the absence of passwords and payment data significantly limits the immediate account-takeover risk, the leaked information still provides valuable material for phishing and social engineering. A scammer could use a real email address, name, country, chess rating, and subscription status to create a convincing fake Chess.com message about a tournament invitation, membership renewal, account-security alert, or fair-play issue. Chess.com users should therefore treat unexpected emails or messages referencing their accounts with extra suspicion and avoid clicking login links; instead, they should go directly to Chess.com through their browser or app.

Read more about it here.

US health tech giant CareCloud started notifying 350,000 people that their information was stolen in a data breach. Hackers accessed one of its AWS environments between March 10 and March 16, and likely exfiltrated data from it.

The initial March 27 disclosure to regulators didn’t provide much details. On June 24, the investigation determined that personal, financial, and medical information was compromised in the incident.

Stolen data included people’s names, postal addresses, Social Security numbers, as well as government-issued identification numbers, such as passports and driver’s licenses. The stolen data included financial information, such as bank account information and payment card numbers, including in some cases the CVV of credit cards, alongside a wealth of medical and health-related information.

Read more about it here.

On May 20, 2026, DentaQuest detected unauthorized access to parts of its network and launched a forensic investigation, which determined that attackers had access between May 17 and May 20, 2026. On June 5, 2026, DentaQuest publicly disclosed the cybersecurity incident, while the ShinyHunters hacking group claimed responsibility, stating it had stolen 234 GB of data after unsuccessful ransom negotiations and had begun leaking the information online.

On July 17, 2026, DentaQuest began mailing notification letters to affected individuals and offered 24 months of complimentary credit monitoring and identity theft protection. DentaQuest has confirmed that at least 15 million individuals were affected, although the total could ultimately exceed 23.4 million as the ongoing data review continues.

The compromised information includes names, addresses, Social Security numbers, member identification numbers, Medicaid numbers, Medicare numbers, and dental or vision health information, including provider names, diagnoses, treatment details, and billing information. Earlier analysis of the leaked files also found email addresses, phone numbers, dates of birth, gender, health insurance enrollment information, government-issued identification numbers, and more than 1.7 million unique Social Security numbers, with some records reportedly dating back to at least 2009.

Read more about it here.

Microsoft Patch Tuesday security updates for July 2026 broke a record: Microsoft provided fixes for 628 CVEs across Windows OS (416 CVE’s), Microsoft Office, Azure cloud services, Defender, Exchange Server, Edge, SharePoint Server and SQL Server. At least three CVEs are exploited in the wild:
CVE-2026-56155 Active Directory Federation Services Elevation of Privilege Vulnerability
CVE-2026-56164 Microsoft SharePoint Server Elevation of Privilege Vulnerability
CVE-2026-58644 Microsoft SharePoint Remote Code Execution Vulnerability

The record-breaking volume of patches signals a significant increase in vulnerability discovery rates across Microsoft’s product ecosystem.

Microsoft cited its use of AI to help uncover previously undiscovered security bugs in its software.

Read more about it here.

Cloud security firm Sysdig has released details on what it claims to be the world’s first ransomware campaign, completely driven by agentic AI, dubbed JadePuffer. The attackers exploited CVE-2025-3248 in an internet-facing Langflow (low-code AI builder tool for agentic applications) instance, allowing the AI agent to autonomously perform reconnaissance, steal credentials, move laterally, and establish persistence without human intervention.

According to Sysdig, the LLM adapted to failures in real time, even correcting unsuccessful login attempts within 31 seconds. The campaign harvested cloud, database and object-storage credentials before compromising a production Alibaba Nacos server and attempting to exploit CVE-2021-29441. Instead of focusing on extortion through recoverable encryption, JadePuffer encrypted all 1,342 Nacos configuration items and deleted the originals, making recovery effectively impossible. Sysdig found that the encryption key was randomly generated, displayed only briefly, and never stored or transmitted, meaning victims could not recover their data even if they paid a ransom. The researchers concluded that the campaign demonstrates how AI agents can autonomously execute complex, adaptive, multi-stage cyberattacks with little or no direct human involvement.

Read more about it here.

LastPass has disclosed that customer data was exposed in a cyberattack targeting Klue, a third-party platform integrated with its Salesforce environment, rather than LastPass’s own systems. The attackers reportedly exploited stolen OAuth tokens to access CRM data, exposing customer names, contact information, support case details, and sales records, but not password vaults, master passwords, or encrypted user data. LastPass said it quickly responded by revoking Klue access, rotating compromised API tokens, investigating the incident with Klue and Salesforce, and notifying law enforcement. The breach also affected several other cybersecurity companies, highlighting the growing risks of supply chain attacks and the abuse of third-party SaaS integrations. Although users’ password vaults remain secure, LastPass is advising customers to stay alert for phishing and social engineering attempts that may use the stolen contact information.

Read more about it here.

Microsoft Patch Tuesday security updates for June 2026 broke a record: Microsoft provided fixes for 208 CVEs across Windows OS, Microsoft Office, Azure cloud services, Exchange Server, Hyper-V, Secure Boot, BitLocker, and a range of AI tools. If we add Chromium and third-party components bundled in Microsoft products, the total reaches 571 CVEs for the month. At least two CVEs are exploited in the wild.

The record-breaking volume of patches signals a significant increase in vulnerability discovery rates across Microsoft’s product ecosystem.

Read more about it here.