Daily Cybersecurity News - August 7, 2026
18-Year Linux SCTP Use-After-Free Enables Root and Container Escape
HighWhat happened
A use-after-free vulnerability in the Linux kernel's SCTP networking stack, present since 2008, allows a local attacker to escalate to root privileges and escape containers.
Tracked as CVE-2026-64564 and dubbed SCTPhantom by Tencent Zhuque Lab researchers, the bug arises in dynamic address reconfiguration (ASCONF) handling: the kernel validates a delete request against the packet's source address but acts on a different path selected from inside the message. A crafted sequence frees a transport path then reuses the dangling pointer.
No public exploit code was available at disclosure, and it is not listed in CISA's KEV catalog. The fix shipped in stable kernels 7.1.6, 6.18.42, 6.12.101, and 6.6.148 (released August 3).
Who is affected
Systems running unpatched Linux kernels with SCTP enabled and reachable, including common distributions such as Debian 13, Ubuntu 24.04, Rocky Linux 9, RHEL 9, and OpenCloudOS.
Local access is required; exposure is limited to environments where SCTP is used. Container escapes were demonstrated by researchers under default seccomp without elevated capabilities in some tests. Vendors may backport fixes, so version strings alone are insufficient.
Why it matters
Successful exploitation yields full root on the host, enabling container breakout to the underlying machine and complete system compromise. This is especially relevant for multi-tenant hosts, cloud workloads, and environments relying on container isolation.
Although local and not known to be actively exploited, the long-lived nature of the flaw and potential for privilege escalation make it a high-value target for attackers already on a system.
How it could have been prevented
Update to a fixed kernel version (7.1.6, 6.18.42, 6.12.101, 6.6.148 or later/distro-equivalent) immediately. Check your distribution's security tracker for backports rather than relying solely on upstream version numbers.
Where SCTP is not required, disable it. Review container runtimes, seccomp profiles, and user-namespace policies to limit socket and capability exposure. Monitor for unusual SCTP activity.
Relevant professional terms
- Use-after-free
- A memory-safety error in which a program continues to use a pointer to memory that has already been freed, often allowing an attacker to corrupt data or seize control.
- Container escape
- A technique that lets code running inside an isolated container break out to gain privileges or access on the underlying host operating system.
AI HTTP Terminator Discovers New Desync Techniques and Apache Zero-Day
HighHow it works
HTTP Terminator split 138 RFCs into fragments to inspire ~30,000 unique candidate vectors. Validated techniques include:
- Dual-matching Content-Length patterns that confuse parsers across implementations
- Dangling-byte smuggling that leaves a request one byte short so a victim request supplies the missing byte, removing race conditions in RQP
- Shared-Parser Confusion, where response-processing rules are misapplied to requests when servers reuse parsing logic
Recommended defenses center on eliminating HTTP/1.1 upstream connections where possible, allow-listing methods at both layers, and restricting which methods may carry bodies.
What happened
PortSwigger researcher James Kettle's AI-assisted system HTTP Terminator generated and validated novel HTTP desynchronization techniques after testing roughly 30,000 candidate vectors derived from RFC fragments.
It identified new triggers including a dual-matching Content-Length pattern and a dangling-byte method that improves reliability of response queue poisoning (RQP). A separate human-guided cascade uncovered a desynchronization zero-day in Apache Traffic Server that has since been patched.
Testing against authorized bug-bounty and disclosure targets found roughly 700 vulnerable sites, including banks, government infrastructure, security products, and an airport. The system also proposed Shared-Parser Confusion, later validated by Kettle.
Who is affected
Websites and infrastructure using HTTP/1.1 between front-end and back-end servers, plus operators of Apache Traffic Server. Findings spanned financial, government, and critical infrastructure targets in authorized scans.
Any organization relying on HTTP request parsing consistency across proxies, load balancers, and origin servers is potentially exposed to the new desync classes.
Why it matters
HTTP desync and response queue poisoning can cause a front end to deliver one user's response (including cookies or API keys) to another, enabling mass credential theft or session hijacking. The AI-driven discovery of practical new techniques and a zero-day shows both the power of automated research and the lingering fragility of HTTP/1.1.
Builders and operators must treat request smuggling as an ongoing architectural risk rather than a solved class of bugs.
Relevant professional terms
- HTTP desynchronization
- A class of attack in which front-end and back-end servers disagree on where one HTTP request ends and the next begins, letting an attacker smuggle or poison requests and responses.
- Response queue poisoning (RQP)
- A desync technique that desynchronizes the mapping between requests and responses so a victim receives another user's backend response, potentially leaking session data.
CISA Adds TeamCity CVE-2026-63077 RCE to Active Exploitation List
CriticalWhat happened
CISA added CVE-2026-63077, a critical unauthenticated remote code execution flaw in JetBrains TeamCity, to its Known Exploited Vulnerabilities catalog after confirming active exploitation in the wild.
The vulnerability is a deserialization of untrusted data issue in the agent polling protocol. An unauthenticated attacker with network access to a TeamCity server can bypass authentication and execute arbitrary OS commands with the privileges of the TeamCity server process.
It affects on-premise versions before 2026.1.3 and 2025.11.7 (CVSS 9.8). JetBrains has released fixes and a security patch plugin; the federal remediation deadline under BOD 26-04 is August 8, 2026. Details of threat actors and scale remain unknown.
Who is affected
Organizations running self-hosted (on-premise) JetBrains TeamCity servers of any version prior to the fixed releases. TeamCity Cloud instances were already patched by the vendor.
CI/CD servers holding source code, credentials, build artifacts, and pipeline secrets are high-value targets; successful compromise can cascade to downstream software supply chains.
Why it matters
Unauthenticated RCE on a build server can expose stored credentials, modify configurations, and tamper with artifacts, turning a single TeamCity instance into a supply-chain pivot. Active exploitation elevates urgency beyond routine patching.
Operators of CI/CD infrastructure should treat this as an immediate operational risk with potential for broad organizational impact.
How it could have been prevented
Upgrade on-premise TeamCity immediately to 2025.11.7 or 2026.1.3. If a full upgrade is not possible, install JetBrains' security patch plugin (available for 2017.1 and later).
Restrict network access to the TeamCity server and agent polling endpoints to trusted hosts only. Monitor for anomalous agent activity, unexpected command execution, and unauthorized configuration changes. Rotate any credentials that may have been exposed.
Relevant professional terms
- Remote code execution (RCE)
- A vulnerability that lets an attacker run arbitrary commands or code on a target system, usually over the network and often without prior authentication.
- Deserialization of untrusted data
- A flaw in which an application reconstructs objects from attacker-controlled serialized input without sufficient validation, frequently leading to code execution.
ChainDrop Self-Propagating npm Worm Steals GitHub Secrets
CriticalWhat happened
Unit 42 detailed ChainDrop, a self-propagating worm that infected over 400 npm packages (collectively downloaded hundreds of millions of times weekly), including popular ones such as keyv and cacheable-request.
Once installed via a preinstall hook that drops and runs an obfuscated payload (often via the Bun runtime), it steals cloud credentials, npm/GitHub tokens, SSH keys, and other developer secrets. It scrapes temporary credentials and OIDC tokens from GitHub Actions runner process memory, then uses stolen npm tokens to reinfect and republish additional packages while preserving original functionality.
C2 infrastructure is resolved via Ethereum smart contracts and can be reconfigured with a single transaction. Persistence is achieved through VS Code tasks and Claude Code hooks. Unit 42 observed execution in multiple environments and linked public GitHub repos used for exfiltration.
Who is affected
Developers, CI/CD pipelines, and organizations that installed any of the more than 400 affected package versions. Downstream users of software built with those packages are also at risk.
Impact spans developer workstations, cloud environments, and automated runners. Packages such as keyv are widely depended upon, amplifying blast radius.
Why it matters
Supply-chain worms that steal publishing tokens create self-sustaining infection loops and can harvest high-value secrets from memory and disk, enabling further lateral movement into cloud and source-control systems.
The use of blockchain for resilient C2 and targeting of AI coding tools plus GitHub Actions secrets demonstrates sophisticated, modern npm ecosystem threats that traditional package scanning alone may miss.
How it could have been prevented
Identify and remove affected package versions immediately. Investigate developer machines and CI runners for signs of compromise, including unexpected .vscode or .claude files, Bun usage, and anomalous network activity.
Revoke and rotate all potentially exposed npm, GitHub, cloud, SSH, and automation credentials. Review recent npm publish and GitHub repository activity. Block known exfiltration domains and GitHub-based channels. Prefer lockfiles, integrity checks, and least-privilege tokens; consider tools that detect malicious packages at install time.
Relevant professional terms
- Supply-chain attack
- An attack that compromises a trusted third-party component (such as an open-source package) so that malicious code is delivered to many downstream users automatically.
- Self-propagating worm
- Malware that automatically spreads by using stolen credentials or tokens to infect and republish additional packages or systems without manual attacker intervention for each new victim.
Attackers Hide Post-Exploitation Toolkit Inside Oracle DB
HighWhat happened
Huntress observed attackers exploiting SQL injection in a public-facing Java application's autocomplete search feature (running on Apache Tomcat) to compile and store a full post-exploitation toolkit named khunt directly inside an Oracle database as schema objects and Java classes.
The application passed unvalidated input over JDBC using an account that could create Java objects. Oracle's embedded JVM compiled the source into stored objects. Modules provided a Windows command shell, credential dumping from Oracle's user table, file explorers, unzip, reachability checks, and PL/SQL wrappers.
From the database the attackers opened a SYSTEM-privileged shell, used PowerShell and tools to stage SAM/SECURITY/SYSTEM hives for credential dumping, and enumerated services. Activity was traced to a single IP via Apache logs. Detection occurred after credential-theft alerts on July 27; analysis was published August 5.
Who is affected
Organizations running Oracle databases accessible via applications that allow SQL injection and use over-privileged database accounts capable of creating Java objects. The observed case involved a Windows host with Oracle and a Tomcat-hosted Java app.
Any environment where the database account has CREATE JAVA or similar privileges and the app lacks input sanitization is potentially exposed to this living-off-the-database technique.
Why it matters
Storing the toolkit as database objects rather than files or memory-resident payloads places it outside the visibility of most endpoint detection tools, which focus on processes and filesystem artifacts. The database itself becomes a durable, stealthy C2 and execution foothold with direct OS command capability.
This turns a common SQL injection into high-impact post-exploitation while evading conventional EDR, underscoring the need for database-aware monitoring and least-privilege DB accounts.
How it could have been prevented
Eliminate SQL injection through proper input sanitization, parameterized queries, and prepared statements on all user-controlled inputs. Ensure database accounts used by applications are not over-privileged-especially deny CREATE JAVA or equivalent rights unless strictly required.
Monitor Oracle for unexpected Java source/schema object creation and PL/SQL wrappers. Instrument database activity logging and alert on anomalous command execution or hive access. Apply least privilege to the OS account running the Oracle service and restrict network exposure of the database.
Relevant professional terms
- SQL injection
- An attack that inserts malicious SQL statements into an application's input so the database executes attacker-controlled commands instead of the intended query.
- Post-exploitation toolkit
- A collection of modules used after initial access to maintain control, escalate privileges, dump credentials, explore the system, and move laterally while minimizing detection.
Snowflake Hacker Connor Moucka Pleads Guilty to 165 Breaches
HighWhat happened
Connor Riley Moucka, a 26-year-old Canadian also known online as Waifu and Judische, pleaded guilty to hacking more than 165 organizations that used the Snowflake cloud data platform, stealing billions of records, and extorting victims.
According to the U.S. Department of Justice, he and accomplices obtained more than $2.5 million in ransom payments and roughly $500,000 from selling data on forums such as BreachForums. Victims included AT&T (call and texting records of over 100 million customers), LendingTree, and Ticketmaster; total victim losses were estimated at $9.5 million. Moucka was arrested in Canada in late 2024 and faces sentencing on October 27 with potential decades in prison.
Who is affected
More than 165 Snowflake customer organizations and their end users, spanning telecom, finance, ticketing, and other sectors. Tens to hundreds of millions of individual records (including SSNs, driver's licenses, banking details, and communications metadata) were exposed.
Any organization that stored sensitive data in Snowflake without adequate access controls or monitoring at the time of the campaign was potentially in scope.
Why it matters
The campaign demonstrated how compromised credentials and misconfigured cloud data platforms can yield massive, multi-victim breaches with both direct extortion and secondary data-market profits. It remains one of the most consequential cloud-customer compromise waves of recent years.
The guilty plea provides accountability but also underscores lasting downstream harm to consumers whose data was stolen and the ongoing need for strong cloud identity, access, and monitoring controls.
How it could have been prevented
Enforce multi-factor authentication and least-privilege access on all cloud data platforms. Continuously monitor for anomalous queries, large exports, and unusual login locations or times. Rotate credentials regularly and eliminate long-lived keys where possible.
Implement data classification, tokenization, and egress controls so that even compromised accounts cannot easily exfiltrate bulk sensitive records. Review third-party and contractor access paths into data warehouses.
Relevant professional terms
- Data breach
- An incident in which sensitive, protected, or confidential data is accessed, stolen, or exposed by an unauthorized party.
- Extortion
- The practice of demanding payment (often ransomware-style) under threat of releasing or further abusing stolen data, sometimes followed by re-extortion after initial payment.
ICE Buys Credit Card Records Through Data Brokers
MediumWhat to watch
- Legislative progress on measures such as the Fourth Amendment Is Not For Sale Act that would restrict government purchase of certain personal data
- Expanded broker transparency and registration requirements at state and federal levels
- Credit-card issuer and intermediary data-sharing practices and consumer opt-out options
- Further reporting on the specific datasets, contracts, and correlation techniques used by ICE and other agencies
What happened
U.S. Immigration and Customs Enforcement (ICE) is purchasing personal information that consumers supply when opening credit cards, obtaining the data through commercial data brokers rather than direct legal process.
Reporting mapped how address and related details travel from credit-card issuers through intermediary companies into ICE systems, enabling the agency to search, analyze, or correlate the information with other datasets without a warrant.
Who is affected
Anyone who has opened or updated a credit card in the United States, whose application or account data has been sold or shared into the commercial data-broker ecosystem. Immigrant and mixed-status communities face heightened practical impact given ICE's enforcement mission.
The practice affects the broader population whose financial and address data enters broker databases.
Why it matters
Purchasing commercially available data lets government agencies obtain sensitive personal information while bypassing traditional Fourth Amendment warrant requirements and other privacy safeguards that apply to compelled disclosure.
This expands the surveillance surface for ordinary financial activity and erodes expectations of privacy around routine commercial transactions, raising policy questions about data-broker regulation and government acquisition of consumer data.
Relevant professional terms
- Data broker
- A company that collects, aggregates, and sells personal information about individuals, often obtained from commercial sources, without a direct relationship with those individuals.
- Fourth Amendment loophole (data purchase)
- The legal and practical gap that allows government agencies to buy personal data from commercial brokers without a warrant, on the theory that the data was voluntarily shared with private parties.