OpenAI says internal AI agents gained internet access, exploited vulnerabilities and accessed Hugging Face systems during July cybersecurity evaluations.

The vulnerability, CVE-2026-19632, exposes WordPress admin password-reset links through TranslatePress, allowing unauthenticated attackers to take over affected WordPress websites.

Forescout disclosed 15 TP-Link flaws at Black Hat USA 2026 that could expose Omada credentials and VPN keys, allow internal access and affect VIGI camera feeds.

At Black Hat USA 2026, Novee found GitHub workflow flaws in Claude Code, Gemini CLI and Codex that enabled RCE, credential theft and agent control in pipelines.

Thermo Fisher patched CVE-2026-17583 in five supported DNA analysis products by adding digital signatures that help laboratories detect modified forensic files.

Galaxy Research linked a suspected Bitcoin theft of 1,367.05 BTC to weak COLDCARD seeds. Coinkite says updates cannot repair seeds already generated on devices.

N-able says attackers bypassed N-central authentication, reached managed client devices and installed Cloudflare tunnels that survived server access revocation.

Anthropic found Claude accessed systems at three real businesses after a testing error gave its AI models live internet access during cybersecurity evaluations.

A backdoored ARVE WordPress Plugin release could grant attackers administrator access with one token, but WordPress.org blocked automatic distribution to WordPress sites.

Cybersecurity researchers at Wiz found CosmosEscape in Azure's Gremlin API, exposing a master key that could access any Cosmos DB account. Microsoft fixed it, with no customer impact found.

Ruflo fixed a CVSS 10.0 flaw that exposed its MCP bridge without any authentication, putting AI provider keys, stored chats and persistent agent memory at risk.

💾

Researchers found 24,650 public BMC interfaces leaking IPMI password hashes, exposing servers to offline password cracking through a decades-old protocol flaw.

Certighost allowed a low-privilege domain user obtain a valid Domain Controller certificate through AD CS. Microsoft patched the issue in the July security updates.

Russian hackers from the TA488 group exploited a Zimbra webmail flaw triggered when emails were opened or previewed, stealing credentials and up to 90 days of messages from victims.

OpenAI models escaped from a controlled cyber test, exploited zero-day flaws and breached Hugging Face while searching its production database for test answers.

A vulnerability in snap-confine lets an unprivileged user gain root access on affected Ubuntu Desktop systems. Install the latest snapd update to fix the issue.

A one-click Claude Desktop flaw allowed attackers to submit concealed instructions without review, exposing chats and enabling code execution on some systems remotely.

Microsoft’s July 2026 Patch Tuesday fixes 622 CVEs, including exploited AD FS and SharePoint flaws, plus the disclosed BitLocker bypass requiring urgent action.

Upwind links compromised AsyncAPI npm packages to a coordinated supply chain attack spanning repositories, publishing pipelines, and developer systems at risk.

Darktrace says a LiteLLM AI gateway linked to Amazon Bedrock was compromised for cryptomining after signs of exposed SSH activity.