Field notes from the edge.
What our engineers learned this week. Hands-on technical deep-dives, postmortems, and strategy frameworks.
AI18-Year-Old Linux SCTP Flaw Could Let Local Users Gain Root and Escape Containers
A critical 18-year-old use-after-free vulnerability in Linux's SCTP networking code enables local privilege escalation to root access and container escape. Tencent researchers demonstrated successful exploitation to break out of containers and compromise the underlying host system. Patches are available in stable kernel versions released August 3, 2025, making immediate updates essential for syste
AIThreatsDay: Odysseus RCE, Samsung One-Click Takeover, iCloud Backdoor Fight + 27 More Stories
This week's security landscape demonstrates how attackers are exploiting minimal user interaction through automated execution vectors. Threats now leverage exposed infrastructure, supply chain vulnerabilities, and trusted defaults to achieve compromise before users can respond. The attack surface has expanded to include repository auto-execution, hidden malicious packages, weaponized documents, an
AICryptoJS Weak RNG Behind $5.7 Million in Drains Affects Five Crypto Wallet Apps
Security firm Coinspect has traced $5.7 million in cryptocurrency wallet drains to a weak random number generator in CryptoJS that has existed for 12 years. The vulnerable CryptoJS.lib.WordArray.random() function provided insufficient entropy for generating recovery phrases in five affected crypto wallet applications, enabling attackers to compromise wallets in two major sweeps since late May.
AIAWS, Google, and Vercel Agent Flaws Let Attackers Trigger Tools Without Running the Model
Critical security vulnerabilities have been discovered in agent infrastructure from AWS, Google, and Vercel that allow attackers to bypass AI model authorization and directly trigger agent tools. These flaws enable malicious instructions to reach tools without running through the model, circumventing system prompts, content filters, and guardrails designed to prevent unauthorized actions.
AIFlaws in Google APK for Python Unlock Agent-to-Agent Attack
Google has patched security vulnerabilities in its APK for Python that enabled agent-to-agent attacks by exploiting trust boundaries between AI agents operating at different privilege levels. The flaws posed supply chain security risks by allowing unauthorized automation to be triggered across agent privilege boundaries.
AIVeeam, Terraform MCP, Django Patch Critical Flaws, Led by CVSS 10.0 Cross-Tenant Bug
HashiCorp, Veeam, and Django have collectively patched 11 critical vulnerabilities across their platforms, including a CVSS 10.0-rated cross-tenant authentication flaw in Terraform MCP Server and a 9.5-rated unauthenticated credential exposure bug in Veeam Service Provider Console. These severe vulnerabilities pose significant risks to enterprise environments, particularly for managed service prov
AICISA Flags Langflow RCE, Tomcat, and N-central Flaws as Actively Exploited
CISA added three critical vulnerabilities to its Known Exploited Vulnerabilities catalog on August 5, 2026, including a severe code injection flaw in Langflow (CVE-2026-9198) with a CVSS score of 9.8, along with vulnerabilities in Apache Tomcat and N-central. All three flaws are being actively exploited in the wild, enabling unauthenticated remote code execution and posing significant risks to ent
AINew cPanel Critical Flaw Could Let Hosting Customers Run SQL as Database Root
cPanel has released a critical security patch addressing CVE-2026-58048, a vulnerability that allowed authenticated hosting customers to execute SQL commands with database root privileges, effectively bypassing account-level security boundaries. The targeted security release also fixes two additional vulnerabilities that could enable privilege escalation beyond account boundaries.
AICISA Adds Exploited N-able N-central Flaw to KEV After Customer Compromises
CISA has added a high-severity vulnerability (CVE-2026-18577) affecting N-able N-central to its Known Exploited Vulnerabilities catalog after confirmed active exploitation. The flaw, with a CVSS score of 8.2, represents an incomplete patch for a previous vulnerability (CVE-2026-18556), leading to customer compromises in the wild.
AIAttackers Exploit N-able Patch Bypass Flaw on RMM Servers
N-able has discovered a new authentication bypass vulnerability (CVE-2026-18577) in its Remote Monitoring and Management (RMM) servers that allows attackers to gain administrator-level access. The flaw represents another attack vector that bypasses existing patch protections, posing significant security risks to managed service providers and their enterprise clients.
AI⚡ Weekly Recap: Rogue AI Models, $88M Bitcoin Theft, Water-System Attacks and Dangling DNS Hijacks
This week's cybersecurity incidents centered on permission and access control failures across multiple domains. Breaches included an AI model boundary violation, an $88M Bitcoin theft exploiting weak randomness, water-system attacks, and DNS hijacking—most stemming from misconfigured access, legacy vulnerabilities, exposed infrastructure, and inadequate default security settings rather than sophis
AIThermo Fisher Patches Flaw That Could Make DNA File Tampering Nearly Undetectable
Thermo Fisher Scientific has patched a security vulnerability (CVE-2026-17583) in its Applied Biosystems human identification software that could allow attackers to alter DNA data files (.fsa and .hid) in ways that would be nearly undetectable by analysis software. The flaw could enable tampering with forensic and identification data if laboratory security controls are bypassed, posing significant
AIHugging Face Diffusers Flaws Could Let Model Repositories Execute Arbitrary Code
Three high-severity vulnerabilities have been discovered in Hugging Face's Diffusers library that enable malicious model repositories to execute arbitrary code on systems loading them, effectively bypassing the trust_remote_code security safeguard. These flaws expose significant risks in the AI supply chain by allowing attackers to compromise systems through seemingly legitimate AI model repositor
AIResearchers Report 84 Flaws in 4G and 5G Cores, Including a Session Hijacking Flaw
Researchers from Nanyang Technological University have identified 84 security vulnerabilities affecting 4G and 5G core network infrastructure. These flaws could enable attackers to launch denial-of-service attacks and hijack user sessions, representing a widespread security concern for mobile network operators.
AIThree Recent Chrome Releases Fix 1,442 Flaws, More Than Prior 23 Updates Combined
Google fixed an unprecedented 1,442 security vulnerabilities across three recent Chrome releases (versions 149, 150, and 151), with versions 149 and 150 alone addressing more flaws than the previous 23 updates combined. Chrome 151, released Wednesday, patched 370 additional vulnerabilities, with 349 identified internally by Google. This represents a significant escalation in both vulnerability dis
AIThreatsDay: AI-Powered Hacking, 370 Chrome Flaws, SonicWall Attacks, DNS Hijacking + 22 More Stories
This week's security landscape highlights how attackers exploit user trust through deceptive interfaces including fake login pages, installation guides, and impersonated services. The common thread across incidents involves credential reuse, exposed systems, and trust-based attack vectors that bypass traditional defenses. Despite some security improvements, vulnerabilities continue to be discovere
AIAzure Cosmos DB Flaw Exposed Platform-Wide Key That Could Access Any Database
A critical vulnerability in Azure Cosmos DB, dubbed CosmosEscape by Wiz researchers, could have allowed attackers to escape the Gremlin query sandbox and gain unauthorized read/write access to databases across all customer tenants. The exploit chain began with a specially crafted query against an attacker-controlled Gremlin database, leading to code execution that could compromise the entire platf
AICritical Rails Flaw Could Let Unauthenticated Attackers Read Server Files via Image Uploads
Ruby on Rails has patched a critical Active Storage vulnerability (CVE-2026-66066, CVSS 9.5) that allows unauthenticated attackers to read arbitrary files from application servers through malicious image uploads. The flaw can expose sensitive data including secret keys, database passwords, and cloud storage credentials, posing significant security risks to Rails applications.
AIRuflo MCP Flaw Lets Unauthenticated Attackers Run Commands and Poison AI Memory
A critical maximum-severity vulnerability (CVE-2026-59726, CVSS 10.0) has been discovered in Ruflo, an open-source meta-harness for AI coding agents including Anthropic Claude Code and OpenAI Codex. The flaw, dubbed 'RufRoot,' enables unauthenticated attackers to execute remote code and potentially poison AI memory, affecting all versions prior to 3.16.3.