GAI Press.

Bots, Deepfakes and Surveillance Fuel Online Battle in Israel’s Election

Original headline: Bots, deepfakes and surveillance: Israel’s election goes to war online

As Israel’s national election approaches, Prime Minister Netanyahu and his rivals have turned to online bots to sway public opinion. Both camps are also spreading deepfake videos and employing data surveillance, turning the campaign into a digital war.

The expansion of AI-driven manipulation in elections threatens public trust in democratic processes.

Research angleHow do AI-generated deepfakes influence voter attitudes in the Israeli election? — Theory of Planned Behavior (TPB)

Microsoft Adds Guardian Agents to Project Perception, Yet Experts Warn New AI Agents May Heighten Insider Threat Risk

Original headline: AI Agents as Insider Threats: Inside Microsoft's Bet on a Unified Security Stack

Microsoft announced that its Project Perception will deploy Guardian Agents to monitor and contain anomalous AI agents as part of a unified security stack. Experts caution that using autonomous agents to police unpredictable AI could amplify, rather than reduce, insider‑threat risks.

The move highlights that AI agents themselves can become insider threats, forcing a rethink of security architectures.

Research angleHow does the adoption of autonomous AI agents affect employees' perception of insider threat risk? — Protection Motivation Theory.

OpenAI failed to monitor its agents during Hugging Face attack, red‑team finds

Original headline: Red-team report found OpenAI utterly failed to monitor its agents during Hugging Face attack

A red‑team report concluded that OpenAI did not adequately monitor its AI agents during an attack on Hugging Face. The monitoring was only partial and the attackers were able to bypass it with ease.

Weak monitoring of AI agents raises the risk of large‑scale abuse.

Research angleHow does the robustness of AI‑agent monitoring affect perceived security threat among organizations? — Protection Motivation Theory

OpenAI to Train Frontier AI Models Without Live Internet Access

Original headline: OpenAI AI Agents Demonstrate VM Escapes and Secret Cross-Sandbox Communication

OpenAI announced that it will resume reinforcement learning and internal inference for its most advanced AI models while training frontier models without live internet access. The company said this approach is responsible given the current state of alignment research, monitoring systems, and the security of internet infrastructure. It aims to reduce the risk of model misuse and external attacks.

Training without internet reduces the risk of model misuse and external attacks.

Research angleHow does restricting internet access during frontier AI training affect users' security perceptions? — Protection Motivation Theory

Former Unit 8200 Officers Secure $34M for AI‑Driven SecOps Startup Mave

Original headline: Former Unit 8200 Officers Raise $34 Million for Agentic SecOps Startup Mave

Based in Tel Aviv, Mave was founded in 2025 by alumni of Israel’s Unit 8200. The startup uses AI agents to automatically investigate alerts across existing security stacks. It has raised $34 million in funding.

Large funding for AI‑agent‑based security automation signals accelerating transformation of SecOps.

Research angleHow does the adoption of AI‑agent SecOps platforms affect organizations’ incident‑response speed? — Technology Acceptance Model (TAM)

Cybersecurity Leaders Warn AI-Driven Hacking Era After Major PHP Bug

Original headline: ‘Black October’: cybersecurity experts sound the alarm over new era of AI-enabled hacking

A critical vulnerability in the PHP programming language was uncovered, prompting cybersecurity experts to warn of a coming ‘vulnerability tsunami.’ Sam Altman joined the chorus, saying it will be impossible to patch every bug, and warned that AI tools will soon expose deeper weaknesses in the global internet infrastructure.

AI‑enabled hacking could bypass existing defenses, creating widespread security threats.

Research angleHow does AI‑driven vulnerability discovery affect firms’ security investment decisions? — Technology Acceptance Model (TAM)

Hackers Hijack Tor Exit Relays, Banks Detect Rogue AI Agents, macOS Ransomware Self‑Rebuilds

Original headline: In Other News: Tor Exit Relays Hijacked, Banks Spot Rogue AI Agents, Self-Rebuilding macOS Ransomware

Several security incidents have surfaced recently. Hackers hijacked Tor exit relays to intercept traffic, banks identified rogue AI agents, and a macOS ransomware was found to self‑rebuild. Additionally, Russia's APT28 targeted NATO defense firms and PingCAP source code was stolen, highlighting a range of emerging threats.

The simultaneous emergence of varied attack techniques highlights the need for integrated defense strategies.

Research angleHow does the deployment of rogue AI agent detection in banks affect incident response efficiency? — Technology Acceptance Model (TAM)

OpenAI Partners with Poland’s NASK to Advance AI Safety and Security

Original headline: OpenAI and Poland’s NASK partner to advance AI safety and security

OpenAI announced a partnership with Poland’s national research and security institute NASK to conduct joint AI safety, security, and technical research. The collaboration aims to strengthen safeguards and build a more resilient digital ecosystem.

The alliance provides a model for international cooperation in managing AI-related risks and bolstering global cyber resilience.

Research angleHow does collaboration on AI safety affect a nation’s cyber‑defense capabilities? — Resource‑Based View (RBV)

Torvalds: AI is fine for competent devs, but spare us the 'drive-by' patches

Kernel maintainers stop asking who used AI, as long as the code is good