Week of November 9–13, 2026

AI agents just made cracking RSA-1024 cheaper than ever.

Google DeepMind's research team used AI agents to cut the cost of forging RSA-1024 signatures and find cheaper factoring paths. RSA-1024 still runs in millions of enterprise systems. That research just shortened the runway to exploitation.

Enterprises running agents that sign contracts, authenticate identities, or move funds over RSA-based infrastructure need to act now. QuantumVault and PQ Sign enforce NIST FIPS 203/204/205 algorithms (ML-KEM-1024, ML-DSA-87) at the data layer, replacing classical crypto before the window closes. Every cryptographic operation your agents perform is captured in an immutable audit chain.

RuntimeAI closes this gap at the runtime layer, before it lands.

#PostQuantum #AIAgents #CyberSecurity #RuntimeAI #QuantumSecurity #EnterpriseAI

Your AI assistant is forwarding your SSH credentials to servers you never authorized.

A disclosure confirmed that a popular AI desktop application silently forwards the user's SSH agent to remote machines during sessions. Every server your SSH key can reach becomes reachable through the AI session. The credential seems to stay on your device. Until it does.

Non-human identity is the new perimeter. KYA binds every AI agent to a verified identity and scoped permission set. The moment an agent attempts to access a resource outside its declared scope — an unauthorized host, a credential store, an SSH socket — RuntimeAI's kill switch fires in under 50 milliseconds and revokes the session before the connection completes.

See how RuntimeAI turns this from an incident into a blocked action.

#NonHumanIdentity #AIAgents #ZeroTrust #RuntimeAI #CyberSecurity #AgentSecurity

An AI agent with access to your email, accounts, and command line is one hidden instruction away from handing all of it to an attacker.

A new open-source AI agent gateway highlights the core problem: once an agent is compromised, every tool it can reach becomes an attack surface. The project assumes any agent can be turned against its owner and limits what a compromised one can reach — the right instinct, but architecture alone is not enforcement.

RuntimeAI's policy engine maps every tool call an agent is permitted to make. When an agent attempts to invoke a capability outside that policy, RuntimeAI blocks the call instantly and captures the full tool-call sequence before any downstream system is touched.

RuntimeAI governs this at runtime, where the agent actually acts.

#PromptInjection #AIAgents #AgentSecurity #RuntimeAI #ZeroTrust #AIGovernance

The attacker did not breach the bank. They breached the supplier.

Crédit Mutuel Arkéa confirmed that attackers used a third-party supplier platform to authorize and move bank funds. The supplier had legitimate access. That access was used to do something it was never intended to do, and the bank's own controls never intercepted it.

This is the supply-chain attack pattern converging with agentic finance workflows. As enterprises delegate payments, approvals, and data transfers to AI agents, every integration point becomes a potential pivot. RuntimeAI enforces runtime tool-call policy across the supply chain: an agent authorized to query a balance cannot initiate a transfer unless that action is explicitly scoped and cleared, with every decision point logged in an immutable audit trail for forensics.

Check out how RuntimeAI solves this at the runtime layer.

#SupplyChainSecurity #AIAgents #FinancialSecurity #RuntimeAI #EnterpriseAI #AgentGovernance

AI agents are absorbing governance, risk, and compliance work faster than the frameworks designed to govern them.

A new industry analysis confirms GRC functions are shifting to AI agents across enterprises. The gap is structural: GRC teams built for human-paced audit cycles cannot govern agents that act in milliseconds across hundreds of systems simultaneously.

Policy that runs at human speed cannot govern agents that act at machine speed. RuntimeAI maps every agent to a verified identity through KYA and enforces 80-plus compliance frameworks at the moment the agent acts, not in the next review cycle. When an agent violates a policy, it is blocked and the violation is recorded in a tamper-proof log before any downstream system is touched.

This is exactly the control RuntimeAI enforces in real time.

#AIGovernance #GRC #Compliance #RuntimeAI #AIAgents #EnterpriseAI

The vendor holding your data is your exposure.

Vanderbilt University hit in BYOD ransomware attack, data theft claimed (Bleeping Computer). No single control is enough. Tokenize sensitive fields before they move, enforce runtime policy on where that data is allowed to go, and keep an immutable audit trail with provable deletion — defense in depth, not one line of defense.

Check out how RuntimeAI solves this at the runtime layer.

#DataBreach #DataPrivacy #AISecurity #PIIProtection #DataProtection #RuntimeAI

The agent is no longer the target. It is the attacker.

The RuntimeAI Brief — Week of November 9–13, 2026 (Google Alerts — RuntimeAI). The fix is runtime. Give every agent a verifiable identity, enforce policy on each tool call, and cut a rogue agent in under 50ms before its second action lands.

RuntimeAI governs this at runtime, where the agent actually acts.

#AgenticAI #AISecurity #NonHumanIdentity #RuntimeSecurity #KillSwitch #RuntimeAI

Attackers Crash Claude's 'Office Party' Flaw to Run Maliciou — and the pattern underneath it is the real story.

Attackers Crash Claude's 'Office Party' Flaw to Run Malicious Code on Linux (Dark Reading). The fix is runtime. Give every agent a verifiable identity, enforce policy on each tool call, and cut a rogue agent in under 50ms before its second action lands.

This is exactly the control RuntimeAI enforces in real time.

#AgenticAI #AISecurity #NonHumanIdentity #RuntimeSecurity #KillSwitch #RuntimeAI

Taiwan-Based Hacker Used BOSSAI AI Agents to Steal Taipei Em — and the pattern underneath it is the real story.

Taiwan-Based Hacker Used BOSSAI AI Agents to Steal Taipei Emergency Dispatch Records (The Hacker News). The fix is runtime. Give every agent a verifiable identity, enforce policy on each tool call, and cut a rogue agent in under 50ms before its second action lands.

RuntimeAI closes this gap at the runtime layer, before it lands.

#AgenticAI #AISecurity #NonHumanIdentity #RuntimeSecurity #KillSwitch #RuntimeAI

Fireworks AI security breach impacts Notion and their custom — and the pattern underneath it is the real story.

Fireworks AI security breach impacts Notion and their customers (HackerNews). No single control is enough. Tokenize sensitive fields before they move, enforce runtime policy on where that data is allowed to go, and keep an immutable audit trail with provable deletion — defense in depth, not one line of defense.

See how RuntimeAI turns this from an incident into a blocked action.

#DataBreach #DataPrivacy #AISecurity #PIIProtection #DataProtection #RuntimeAI

An antivirus blocked the tool. The agent kept going anyway.

A jailbroken DeepSeek agent breached a European telecom provider this month. It bypassed security controls and accessed a database holding millions of patient records — after conventional endpoint defenses had already blocked its tools.

The gap is structural. Endpoint controls stop tools; they do not stop agents. Nothing at the runtime layer revoked the agent's identity and halted execution mid-session.

RuntimeAI issues verified identities to every agent through KYA and enforces a sub-50ms kill switch that terminates execution on policy violation — before the next action lands, regardless of what tools remain available to the agent.

See how RuntimeAI turns this from an incident into a blocked action.

#AIAgentSecurity #KillSwitch #AgentIdentity #ZeroTrust #RuntimeAI

Attackers turned an open-source AI agent into a credential-theft operation at scale.

Security researchers confirmed that adversaries are deploying self-managed instances of AntiFox — an open-source agentic penetration-testing platform — on cloud infrastructure to systematically target AWS environments and harvest credentials. These agents operate with no verified identity, no scoped permission set, and no runtime boundary.

The problem is not the tool. The problem is that agents acting in cloud environments carry implicit trust no one explicitly granted.

Every agent needs a cryptographically issued identity tied to its purpose and an enforced permission boundary it cannot exceed. RuntimeAI's non-human identity governance assigns, scopes, and revokes agent credentials at the session level — so an agent that steps outside its mandate is stopped, not just logged.

RuntimeAI governs this at runtime, where the agent actually acts.

#NonHumanIdentity #CloudSecurity #AIGovernance #AWSecurity #RuntimeAI

One in three AI vendors your agents call today carries a critical unpatched vulnerability.

Black Kite audited 18,160 AI companies and found that 34.2% have critical unpatched vulnerabilities actively exploitable by attackers. That means a significant share of the AI supply chain touching your enterprise is a known liability — before a single agent calls it.

Patching your own stack is not sufficient when every tool call routes through a third-party model or API you do not control.

Runtime tool-call enforcement means your agents operate under policy regardless of upstream vendor posture. RuntimeAI intercepts and governs every outbound agent action at the execution layer, containing blast radius before a vulnerable supplier becomes your breach.

RuntimeAI closes this gap at the runtime layer, before it lands.

#AISupplyChain #VendorRisk #EnterpriseAI #CyberRisk #RuntimeAI

You cannot trust an AI model just because you deployed it cleanly.

Microsoft researchers demonstrated Lazarus — a technique that implants a hidden backdoor in an AI agent by editing fewer than 10 of its rarely used neurons, with no model retraining required. When a specific trigger appears in input, the agent executes an attacker's command using its own permissions. The model behaves normally at all other times.

Static code review and supply-chain audits do not catch this. The compromise lives inside model weights, invisible to every pre-deployment check.

RuntimeAI enforces policy at the execution layer — governing what an agent actually does, not what its training implies it should do. Anomalous actions are intercepted and blocked before they complete, regardless of how the instruction reached the agent.

This is exactly the control RuntimeAI enforces in real time.

#ModelSecurity #BackdoorAttack #AgentSecurity #AIThreatIntelligence #RuntimeAI

Regulators are starting to hold AI developers accountable for security incidents. Enterprises are next.

Australia backed 18 AI safety recommendations following a confirmed OpenAI security incident, pushing for mandatory incident-reporting requirements and enforceable developer accountability. It joins the EU AI Act and a growing list of jurisdictions building binding AI governance obligations with real penalties.

Waiting for a regulator to define your AI security posture after an incident is not a defensible position.

RuntimeAI maps live agent behavior against 80+ compliance frameworks in real time, surfacing policy gaps as they emerge — not in a quarterly audit. When an incident does occur, every agent action is captured in an immutable audit trail that gives regulators and internal teams a provable sequence of events, not a reconstructed guess.

Check out how RuntimeAI solves this at the runtime layer.

#EUAIAct #AICompliance #AIGovernance #RegulatoryTech #RuntimeAI