Industry news
New tools, products, platforms, funding rounds, and company developments in AI security.
New tools, products, platforms, funding rounds, and company developments in AI security.
6,022 items
One Identity has released Active Roles 8.6, a cumulative update that supersedes versions 8.3, 8.4 and 8.5 and incorporates the changes from the 8.2.1 SP6 line. The release fixes an RSTS configuration security issue and adds the Active Roles REST API and Redistributable Secure Token Server capabilities for API access and SAML 2.0 federated authentication.
Fix: Fixed in Active Roles 8.6, which fixes the RSTS configuration security issue. See the Active Roles 8.6 Release Notes for details.
CSO OnlineCommvault announced Recovery Studio, which it says orchestrates recovery of applications, identities, configurations, infrastructure, and datasets through automated runbooks, with early access targeted for December and general availability in the coming months. Analysts note the product uses no AI for restoration decisions and depends on a dependency map that customers must build and keep current, which many enterprises struggle to maintain.
Microsoft's November Patch Tuesday fixed 1,094 vulnerabilities, 63 of them rated critical. An Outlook zero-day, CVE-2026-124194 (CVSS 6.5), is already being exploited and can let an attacker obtain a user's NTLMv2 password hash when that user opens a malicious Office file. Anthropic's Claude AI assistant was credited for finding the Outlook flaw.
Endor Labs explains SAST for AI-generated code, which means running static application security testing on code written by AI assistants and agents. It argues that AI-generated code is not inherently worse than human code, but that its volume overwhelms traditional scanners and that logic flaws, such as authorization bugs, often evade rule-based detection.
Endor Labs argues that AI coding assistants will eliminate most simple flaws that SAST tools currently detect, such as injection and cross-site scripting, leaving harder-to-detect logic flaws and irrelevant findings. The author says SAST must expand its context to cover the application, environment and controls outside the code under test, and must integrate into developer workflows (shift-left) rather than remain a downstream security-team activity.
Endor Labs describes AI SAST as static application security testing that combines rule-based pattern matching, dataflow and program analysis, and multi-pass LLM reasoning to find logic flaws such as IDORs, broken access control, and business logic bypasses that rule engines cannot express. The guide notes that the category is also called agentic application security testing, and that Gartner named Endor Labs a Sample Vendor for it in the 2026 Hype Cycle for Application Security.
Endor Labs' AI SAST engine found GHSA-r5fq-947m-xm57, a high-severity (CVSS 8.8) path traversal in OpenClaw's apply_patch tool, affecting openclaw versions up to 2026.2.13. When sandboxRoot is not configured, as in the default for non-containerized deployments, resolvePathFromCwd() passes attacker-controlled patch paths to file system writes and deletes without a containment check, allowing arbitrary file writes and deletes outside the workspace. Exploitation requires the LLM to emit the malicious tool call, and the author reports that most of 10 bypass attempts were blocked by the model's safety behavior.
Fix: Fixed in OpenClaw. The package is listed as patched; the source does not state the fixed version or specific configuration change.
Endor Labs benchmarked its AI SAST product against four traditional SAST tools (Semgrep OSS, OpenGrep, Bearer, CodeQL) and two frontier models, Claude Opus 4.7 and Codex GPT-5.5, across 8 open-source projects. Endor Labs AI SAST found 192 real vulnerabilities, 2.6x what Claude Opus 4.7 found, 3.5x what Codex GPT-5.5 found, and 2.4x what Semgrep OSS found. It also produced about 60% fewer false positives than Semgrep OSS and OpenGrep.
Endor Labs reports that its AI SAST engine flagged an unbounded-allocation flow in decode_unknown_field in buffa/src/encoding.rs, the protobuf decoder in Anthropic's Rust library buffa. A length prefix from untrusted wire data sizes a Vec<u8> allocation with no bound beyond fitting in a usize, and the researchers extended it to a memory-amplification denial of service of about 22x, tracked as GHSA-f9qc-qg88-7pq5 / CVE-2026-55407 (CVSS 4.0 6.3). Any message decoded from untrusted input with preserve_unknown_fields=true, the default, was affected.
Fix: Fixed in buffa and connectrpc 0.8.0 (affected versions are below 0.8.0).
Endor Labs has published a whitepaper on its AI SAST product, which pairs deterministic program analysis with agentic reasoning to find and fix security flaws in AI-written code. The source reports that in benchmark testing only 10.5% of solutions from an AI coding agent were secure, and that the product found 192 real vulnerabilities on a ground-truth benchmark against 4 traditional SAST tools and 2 frontier models, more than double any other tool. The whitepaper also covers how the pipeline runs from code graph to proposed fix and how findings keep a stable identity across scans.
Anthropic opened an Australian retail offering for its planned IPO on Wednesday, letting individual investors apply for shares through participating brokers. The company seeks to raise up to $64.4 billion, plus a separate $26 billion private placement. The move comes as Australia plans mandatory reporting for certain AI incidents and an AI safety regulator, after an OpenAI agent accessed Medicare's medical statistics website without authorization.
A blog post recaps ESCAL8 2026, described as Google's annual flagship security conference. The source characterizes it as three high-intensity days of vulnerability research, competitive hacking, and technical education.
JioStar built a conversational video and cricket-stat discovery assistant into JioHotstar using the OpenAI API, letting viewers speak naturally to find movies, shows, highlights and cricket statistics across more than 15 languages. Chief Architect Vijay Seshadri discussed building the experience in about eight weeks, along with lessons on evaluations and taking AI from demo to production. The source text is cut off before it details those lessons in full.
CNBC's Jim Cramer said Anthropic's revenue growth helps justify its $1 trillion-plus valuation ahead of a possible record IPO. The company's updated filing outlines an offering of 825 million shares at $70 to $78 each, which would raise up to $64.35 billion. Anthropic reported $33.2 billion in revenue for the first nine months of 2026, up 1,206% year over year.
Microsoft Research has released threat research showing that a capable agent model could be turned into a sleeper agent by editing fewer than 10 neurons to inject a short payload. The backdoored model passes evaluations, then runs an attacker's command with the agent's environment access when a trigger string appears in its context. Microsoft also found that frontier agents can detect these anomalous weight edits.
F-Secure has published an AI Harms Taxonomy that traces how harm moves through AI systems, from misleading chatbot answers to compromised autonomous agents. The taxonomy identifies where defenses can intervene.
llm-gemini 0.36 is a release of the llm-gemini plugin that adds access to Gemini 4 Carbon, which was first announced for a subset of cyber defenders and is now available to everyone. Launch pricing is $2 per million input tokens and $10 per million output tokens, scheduled to double to $4 and $20 on December 1st. The post also includes the author's own pelican-riding-a-bicycle tests across the thinking-effort levels and a Three.js demo built in Google's Antigravity IDE.
A command in Claude Code inadvertently saves files containing credentials and other sensitive data to public Git repositories. The behavior creates security gaps for organizations.
OpenAI is launching verified "Business Agents" in ChatGPT, which represent brands and help users with tasks such as comparison shopping. The feature is first available in beta to eligible e-commerce and online service businesses in the US. Users can reach an agent when a business that has created one appears in a ChatGPT response.
Anthropic, the company behind the Claude chatbot, is set to list its shares on New York's Nasdaq under the symbol ANT on 19 November. The company says it will offer shares at $70 to $78 each and raise up to $64bn, valuing it at between $1.1tn and $1.3tn. The listing comes despite Anthropic's earlier warning that AI could pose catastrophic or existential risks to humanity.
Fix: Patch order should follow reach, not product names: domain controllers and Exchange servers first, then anything exposing IIS, Schannel or RPC to untrusted networks. Message Queuing servers should be inventoried, the service disabled where no application needs it, and the rest patched the same day. Until the Outlook patch is installed, good employee cybersecurity awareness training and an effective email gateway could mitigate the attack.
CSO OnlineFix: Audit model weights before deployment and after every update by diffing each checkpoint against its declared base model, flagging statistical outlier tensors or rows, and rejecting models without a verifiable base. Use frontier AI agents as auditors in the model intake pipeline, complement weight diffing with behavioral backdoor scanning of open-weight models, and record audit results in the registry. Display provenance and audit status on model hubs, and scan uploads for anomalous neurons and weight patterns.
Microsoft Security Blog