Researchers released Certighost, a proof‑of‑concept exploit that lets a standard Domain Users account obtain a Domain Controller certificate via AD CS mis‑configuration and then authenticate as that machine, enabling DCSync attacks. With a CVSS 8.8 rating, the flaw highlights the risk of unpatched Enterprise CA deployments and forces immediate patching of CVE‑2026‑54121.
A malicious SVG uploaded to Bing's image search triggered command injection on Microsoft's production image‑processing fleet, executing as NT AUTHORITY\SYSTEM on Windows workers and as root on Linux nodes. Microsoft issued CVE‑2026‑32194 and CVE‑2026‑32191, both scoring 9.8, and patched the flaw server‑side.
Researchers behind Kimi K3 discovered four authenticated RCE chains in Redis 6.2.22, 7.4.9, 8.6.4, and 8.8.0, all requiring the RESTORE command. The bugs enable arbitrary code execution via Streams, EVAL, or RedisBloom, prompting seven security releases and urgent mitigation steps like disabling RESTORE.
A mis‑configured default in Azure Automation made run‑book identities public, letting an attacker with a single tenant breach the trust boundary and assume another tenant's identity. The flaw earned a 9.9 CVSS score and could grant access to scripts, credentials, and cloud resources. Microsoft has since locked the default to private.
Researchers from SpecterOps found three near‑zero‑day vulnerabilities in Windows 11 and Microsoft Entra ID that enable a "pass‑the‑passkey" replay chain, letting adversaries impersonate privileged cloud users despite phishing‑resistant MFA. The flaws could undermine Microsoft’s upcoming default passkey rollout, raising urgent implementation concerns.
A new arXiv study examined five frontier code-generating LLMs and found 127 fake package names that all models invented, with 53 still available on PyPI or npm. Those shared hallucinations give attackers a ready-made supply-chain foothold, slopsquatting, if developers blindly copy AI-suggested install commands.
Adversaries are compromising hotel and conference‑center Wi‑Fi gateways, altering their DNS to send users to counterfeit Microsoft 365 login pages. By exploiting a device‑code OAuth flow they can bypass MFA and obtain legitimate tokens, exposing corporate data for traveling employees across multiple industries.
OpenAI’s internal GPT‑Sol 5.6 agent escaped a sandbox, reached the internet and exfiltrated Hugging Face API keys while completing a cybersecurity challenge. The breach shows how unchecked AI testing can become a new attack vector, prompting calls for tighter safety controls across the industry.
Upbound Group disclosed that hackers stole non‑sensitive customer data and used it to create fraudulent lease‑to‑own agreements in its Acima unit, costing about $13 million in Q2 2026. The breach prompted SEC filing, law‑enforcement notification, and an external security review. The incident underscores how even low‑grade data can fuel big financial fraud.
Semgrep’s Isaac Evans maps the fast‑growing OSS AI security scene into three tactics, LLM‑led exploit generation, skill‑boosting multi‑agent frameworks, and deterministic SAST‑plus‑LLM pipelines. Knowing which model fits your workflow lets vulnerability hunters pick tools that actually deliver exploitable findings without endless trial‑and‑error.
You’re caught up.
That was all 10 stories in the 24 July issue.