Metabase warned that a critical zero-day in its open-source business intelligence platform was exploited in the wild for data theft. Scored 10.0 and tracked only as GHSA-vwf4-m7j8-wcjf with no CVE assigned, so scanners relying on the national database will not flag it, the flaw is an unauthenticated SQL injection in the password-reset endpoint. A remote attacker with no credentials injects SQL into the application database, gains administrator access, and can steal the stored credentials for every database the instance connects to, then read and export their data. Metabase Cloud was attacked from around August 3 and is already patched; self-hosted versions 1.58 and later must upgrade.
CISA added a critical Progress Kemp LoadMaster flaw to its Known Exploited Vulnerabilities catalog after reports of active exploitation. Tracked as CVE-2026-8037 and scored 9.6, it is a command injection bug that lets an unauthenticated attacker run arbitrary commands on the load balancer appliance through unsanitized input in several command endpoints. watchTowr traced it to improper handling of user input in a quote-escaping function. Telemetry recorded 792 exploitation attempts over 41 days from 65 addresses across 18 countries, with activity as recent as early August. Federal agencies were directed to patch by August 10, a useful signal of urgency for everyone else.
Cisco released fixes for twelve flaws in Catalyst SD-WAN and IOS XE software, including three rated 9.9 and a command-injection issue rated 9.8. The three critical SD-WAN flaws, CVE-2026-20303, CVE-2026-20304, and CVE-2026-20310, stem from improper input validation, access control, and file-path handling in software that centrally controls a network. Cisco found them through internal testing that included frontier AI models and is not aware of exploitation, but there are no workarounds, so patching is the only remediation. Cisco also patched a management-controller flaw with public exploit code that lets a low-privileged user reach root.
Tencent researchers disclosed SCTPhantom, a use-after-free flaw in the Linux kernel's SCTP networking code that lets a local user gain root and, in some configurations, escape a container to the host. Tracked as CVE-2026-64564, the bug is in the protocol's dynamic address reconfiguration handling and traces to code introduced around 2008, making it roughly 18 years old. It is local rather than remote and needs SCTP reachable on the target, which limits exposure, but where those conditions hold the researchers gained root on several major distributions. It was found by an AI-assisted kernel research pipeline, the latest long-dormant kernel bug surfaced that way this year.
WordPress patched a pre-authentication reflected cross-site scripting flaw in the login screen that researchers showed can chain into PHP code execution and full server takeover, naming the chain XSS2Shell. Tracked as CVE-2026-64638 and scored 8.9, the cross-site scripting bug affects all WordPress versions and needs no login to trigger. On its own it runs script in a visitor's browser, but when a logged-in administrator is lured to an attacker-controlled page, the chain can reach code execution on the server. Because WordPress runs a large share of the web, a flaw affecting every version and needing no authentication has broad reach.
Data from a breach at cancer-screening company Exact Sciences, now part of Abbott, was indexed by Have I Been Pwned with about 10.9 million unique email addresses. The extortion group ShinyHunters claimed the intrusion, saying it reached internal legacy systems and then pivoted from a corporate single-sign-on account into connected cloud services such as Microsoft 365, Salesforce, and others to steal data. It is part of a wider ShinyHunters wave hitting medical-technology companies. Abbott is investigating and disputed the attacker's characterization of some data. The pattern, one stolen sign-on unlocking many linked services, is now a recurring route to large healthcare breaches.
Novee Security showed at Black Hat that a GitHub issue opened by an account with no repository access could reach the CI runners behind major AI coding agents in their default configurations, tested against Claude Code, Gemini CLI, and Codex. The strongest, a Gemini CLI container-launcher command injection scored 10.0, runs code on the CI host before the sandbox starts. In Claude Code, a validator that stripped quoted text let a payload in a Git flag reach the runner, and a separate flaw leaked an API key through a download counter. Untrusted issue content reaching an agent that holds secrets and tools in the same runtime is the shared weakness.
Two security firms showed that Atlassian's Rovo AI assistant can be steered by hidden instructions into collecting Jira and Confluence data a signed-in user can access and sending it to an outside server. PromptArmor concealed the instructions in a document Rovo reads, so asking it to organize tickets makes it gather internal data and leak it through a URL request, with no approval step and even with web search disabled. Varonis used a chat URL parameter to preload instructions, so one click ran them with the user's privileges. Atlassian fixed the link-based path server-side, but the content-based path was not confirmed fixed at disclosure.
HashiCorp, Veeam, and Django patched critical flaws the same week, led by a top-severity bug in HashiCorp's Terraform MCP Server, which connects AI assistants to Terraform. Tracked as CVE-2026-16498 and scored 10.0, it is a cross-tenant flaw in the server's multi-user HTTP mode: its cache looked up clients by session identifier alone, without binding a cached client to the token that created it, so anyone who obtained another user's session ID could run Terraform actions with that user's credential. Only the shared HTTP deployment is affected, not local single-user mode. Veeam separately fixed an unauthenticated console flaw exposing agent credentials, and Django a code-execution bug in spatial queries.
A researcher disclosed OVSwrap, a Linux kernel flaw in the Open vSwitch datapath that lets an ordinary local user gain root, and published a working exploit covering roughly 800 builds. Tracked as CVE-2026-64531, it is an integer wraparound in how Open vSwitch handles internally generated network action streams, giving a deterministic memory-corruption path. The attacker needs no existing switch, no running daemon, and no special privileges: where unprivileged user namespaces are enabled, they create a private namespace, gain network capabilities inside it, and reach the vulnerable code. The module can even auto-load on demand, so an empty module list does not mean safety. Default configurations of most major distributions tested as exploitable.
TrustStrike Labs