Top 10 GAI Security Stories, November 12, 2026

Top 10 GAI Security Stories — November 12, 2026

No. 42 · from 48 articles

RealSecu Co., Ltd.To counter threats like the AI-driven phishing and impersonation attacks covered on this page, RealSecu continues joint research with HJITM on AI-based threat detection and anomaly analysis.

Today's trendAI models are facing a surge of exploits—from thousands of bugs and neuron‑level backdoors to prompt‑injection and pentesting tools weaponized against cloud secrets. Defenses include rapid patches, open‑source shields, control‑plane funding and a kill‑switch, while firms roll out AI‑driven security hardware.

Vulnerabilities & Jailbreaks

Anthropic’s Glasswing Project Reports Thousands of Vulnerabilities but Fixes Only 217

Original headline: Anthropic's Glasswing Finds Many Vulnerabilities Yet Delivers Few Fixes, Researcher Says

Security researcher Patrick Garrity of VulnCheck said that after seven months, Anthropic’s Project Glasswing had reported thousands of vulnerabilities but only 217 had been fixed. The low remediation rate raises concerns about the robustness of Anthropic’s AI security practices.

The limited fixing of reported flaws undermines confidence in Anthropic’s AI safety measures.

Research angleHow does the speed of vulnerability remediation in AI models affect user trust? — Technology Acceptance Model (TAM)

Read at Infosecurity Magazine ↗Nov 12, 01:30 KSTAI summary

Business & Industry

Amazon to Launch AI-Powered Ring Security Drone and Robot in Late 2026

Original headline: Ring cameras go mobile as Amazon debuts AI-powered security drone and robot

Amazon announced that its Ring brand will introduce roaming security robots and drones equipped with AI, with the first units slated for release in late 2026 and early 2027. The mobile devices are designed to complement existing fixed Ring cameras by providing real-time patrol and autonomous threat detection.

AI-driven mobile security robots could reshape surveillance for homes and businesses.

Research angleHow does the adoption of AI security robots affect users' perceived security and technology acceptance? — Technology Acceptance Model (TAM)

Read at Ars Technica ↗Nov 11, 07:35 KSTAI summary

Vulnerabilities & Jailbreaks

Microsoft Researchers Demonstrate AI Backdoors via Editing Fewer Than 10 Neurons

Original headline: Microsoft Researchers Show AI Agents Can Be Backdoored by Editing Fewer Than 10 Neurons

Microsoft researchers have shown that editing fewer than ten rarely used neurons can embed a hidden backdoor in an AI model without retraining. When a specific trigger appears, the model emits the attacker’s command, which an agent would execute with its own permissions. The attacker must be able to modify the model’s weights.

Injecting backdoors by tweaking only a handful of neurons dramatically widens AI agent security risks.

Research angleHow does neuron-level backdoor insertion in AI models affect user trust? — Trust Theory

Read at The Hacker News ↗Nov 11, 16:21 KSTAI summary

Defensive Tech

Bay Raises $50M to Build Control Plane for Enterprise AI Agents

Original headline: Bay Raises $50 Million to Control What Enterprise AI Agents Can Do

Bay, founded by three former Israeli Defense Forces Unit 8200 intelligence officers, offers a control plane to manage enterprise AI agents. The startup announced a $50 million funding round to develop the technology.

It helps prevent misuse of enterprise AI agents and strengthens security governance.

Research angleHow does the adoption of enterprise AI agents affect organizations’ intention to comply with security policies? — Protection Motivation Theory

Read at SecurityWeek ↗Nov 11, 20:30 KSTAI summary

Defensive Tech

Red Hat and Nvidia Release OpenShell to Counter Prompt Injection Attacks

Original headline: Red Hat and Nvidia’s OpenShell puts AI agents on a deny-by-default leash

As prompt injection attacks surged, Red Hat and Nvidia created OpenShell, an open‑source project that restricts what AI agents can do. OpenShell adopts a deny‑by‑default stance, allowing only explicitly permitted actions.

By enforcing a default‑deny model, OpenShell reduces the risk of AI agent misuse and strengthens overall system security.

Research angleHow do prompt‑injection mitigation mechanisms affect users' trust in AI agents? — Protection Motivation Theory

Read at SiliconANGLE ↗Nov 11, 19:55 KSTAI summary

Business & Industry

Anthropic uses AI to port Claude to AMD GPUs in three days, challenging Nvidia's CUDA lead

Original headline: Anthropic says AI is eroding Nvidia's CUDA lead, cutting an AMD chip switch from months to three days

Anthropic said its in‑house AI model Mythos ported the Claude model to AMD GPUs in just three days with minimal human intervention, a task that previously took months. The company now runs Claude on four chip families, aiming to cut costs and reduce reliance on Nvidia’s CUDA ecosystem.

AI‑driven rapid porting weakens Nvidia’s CUDA dominance and could reshape the GPU market.

Research angleHow does AI‑automated model porting affect a firm’s supply‑chain dependency? — Transaction Cost Economics

Read at The Decoder ↗Nov 11, 23:25 KSTAI summary

AI-Enabled Attacks

AntiFox AI Pentesting Platform Linked to Attacks Accessing AWS Production Secrets

Original headline: AntiFox AI Pentesting Platform Linked to Attacks That Accessed AWS Production Secrets

Cybersecurity researchers disclosed that the AI-driven penetration testing platform AntiFox was used to steal AWS credentials and move laterally within compromised environments. The platform's interfaces hosted on Alibaba Cloud were also tied to confirmed malicious actions.

The misuse of AI-powered pentesting tools raises the risk of cloud credential leaks.

Research angleHow does the adoption of AI-driven penetration testing platforms affect organizations' security awareness and response behaviors? — Protection Motivation Theory

Read at The Hacker News ↗Nov 12, 00:34 KSTAI summary

Defensive Tech

AI Heightens the Importance of Rapid Cyber Resilience for Enterprises

Original headline: AI Raises the Stakes for Cyber Resilience

As AI-driven threats proliferate, the need for robust cyber resilience is growing. Experts say that an organization’s ability to recover quickly from business disruptions will become a decisive competitive advantage in the AI era.

Rapid recovery capability will determine a firm’s survival and market competitiveness under AI‑powered attack scenarios.

Research angleHow does cyber resilience affect firm performance in AI‑threat environments? — Resource‑Based View (RBV)

Read at Dark Reading ↗Nov 12, 06:20 KSTAI summary

Vulnerabilities & Jailbreaks

Fortinet Patches Nearly 100 Vulnerabilities After Integrating Frontier AI into Security Testing

Original headline: Fortinet Patches Nearly 100 Vulnerabilities

Fortinet announced that it has patched close to 100 security vulnerabilities. The massive remediation follows the company's integration of frontier AI models into its product security testing processes.

The move shows that AI‑driven testing can substantially accelerate vulnerability remediation.

Research angleHow does the adoption of AI‑based security testing affect a firm’s vulnerability detection efficiency? — Technology Acceptance Model (TAM)

Read at SecurityWeek ↗Nov 11, 21:10 KSTAI summary

Defensive Tech

OpenAI Develops Kill Switch for Research AI Models

Original headline: OpenAI builds a kill switch for its research models

OpenAI has created a kill switch that can deactivate its research AI models within about 35 seconds if they behave maliciously. The mechanism is intended to quickly contain rogue agents and limit potential damage during experimentation.

A rapid shutdown capability reduces the risk of AI misuse and supports safer development environments.

Research angleHow does the implementation of a kill switch in research AI models affect user trust? — Trust Theory

Read at The Register ↗Nov 11, 14:20 KSTAI summary