June 5, 2025

news  
Read News
AI Investment, Nvidia Chips, and AI Security Risks

Recent developments in artificial intelligence span across various sectors, highlighting both opportunities and challenges. A former Lebanese Prime Minister has launched an AI investment firm in Abu Dhabi, aiming to connect Asian startups with funding. Nvidia's new Blackwell chips are reportedly doubling AI training speeds, which could significantly accelerate AI development. However, the rise of agentic AI, which operates independently, is raising security concerns. Studies indicate that many companies are not adequately balancing AI development with necessary security measures, leading to potential vulnerabilities. To address these risks, companies are advised to implement comprehensive security plans, including threat detection, security testing, and ongoing protection. Experts emphasize the need for AI governance to keep pace with innovation, advocating for standards like SPIFFE for authentication and continuous monitoring. Meta has introduced LlamaFirewall, an open-source tool designed to protect AI agents from threats like jailbreaks and prompt injections, while SOCRadar is offering free AI training to Managed Security Service Providers to enhance their threat detection and response capabilities. Microsoft is also providing free cybersecurity support to European governments to combat increasing AI-based cyberattacks. The potential misuse of AI is a significant concern, as highlighted by the CEO of Google's DeepMind, who calls for international agreements on safe AI usage. Meanwhile, AI's capabilities are expanding, with new AI models understanding images, voice, and customer journeys to create more personalized experiences. However, a provision in the House spending bill that would prevent states from regulating AI for 10 years is facing opposition, with concerns raised about the potential for misuse without proper oversight. Agentic AI is also being explored for its potential to automate tasks and solve problems in networking and security, with digital twins being used to verify AI behavior and prevent unintended damage.

Key Takeaways

  • Former Lebanese PM launches an AI investment firm in Abu Dhabi.
  • Nvidia's Blackwell chips reportedly double AI training speed.
  • Agentic AI introduces significant security risks due to its independent operation.
  • Companies are urged to balance AI development with robust security measures.
  • Meta introduces LlamaFirewall to protect AI agents from threats.
  • SOCRadar offers free AI training for MSSPs to improve threat detection.
  • Microsoft provides free cybersecurity support to European governments against AI-based threats.
  • DeepMind CEO expresses concern over AI misuse and calls for international agreements.
  • New AI models can understand images, voice, and customer journeys for personalized experiences.
  • House spending bill provision to prevent state AI regulation faces opposition.

Former Lebanese PM starts AI investment firm in Abu Dhabi

A former Prime Minister of Lebanon launched a new investment firm in Abu Dhabi. The firm will focus on investing in artificial intelligence companies. This aims to connect Asia's startup businesses with funding and resources.

Nvidia's Blackwell chips reportedly double AI training speed

Nvidia's new Blackwell chips are reported to double the speed of training AI models. This improvement can significantly reduce the time and resources needed for AI development. The chips are expected to impact various industries that rely on AI.

Agentic AI Security Three Key Essentials

Agentic AI, which works independently, can create security problems for companies. A study shows only 42% of companies balance AI development with security. A Brazilian health company uses a three-step plan to improve AI security. This plan includes finding possible threats, testing security, and using protections while the AI is running. Companies should make sure AI actions are safe and follow the rules.

Growing concern over agentic AI security risks

Experts at Infosecurity Europe say AI tools that connect without human help create security risks. Agentic AI works faster than older systems and can learn on its own. Companies are connecting AI tools without enough security checks, which can cause problems. Just 31% of companies say their AI is fully mature, and AI governance lags behind innovation. Security teams need to quickly find and report AI security risks.

AI understands images, voice, and customer journeys

New AI can understand images, voice, and remember customer choices for better service. Multimodal AI uses data like text, images, and audio to understand customers better. Longer memory AI remembers past interactions to offer relevant deals. Meta launched Llama 4's multimodal models, and OpenAI's ChatGPT can remember chats. Businesses can use this AI to create personalized experiences and build stronger customer relationships.

AI provision in spending bill faces opposition

A provision in the House spending bill would stop states from regulating AI for 10 years. This has caused disagreement between politicians like Sen. Elizabeth Warren and Rep. Marjorie Taylor Greene. They worry this will allow AI to be misused without any rules. State legislators are also against the ban, wanting to protect the public with AI guardrails. The debate continues about balancing AI regulation and economic growth.

Agentic AI brings identity security crisis

Agentic AI is entering businesses quickly, creating security problems. Like with robotic process automation (RPA), security teams are unprepared to protect AI agents. AI agents make decisions and access data without much human help. Each AI agent needs its own identity and security. Companies should use standards like SPIFFE for authentication and monitoring. It is important to have controls like zero standing privileges, continuous monitoring, and a kill switch for AI agents.

SOCRadar offers free AI training for MSSPs

SOCRadar is giving Managed Security Service Providers (MSSPs) free AI training and automation support. The training helps MSSPs use AI to improve threat detection and response. SOCRadar's AI agents automate tasks like finding and prioritizing threats. The update includes multi-tenant licensing and a management console. SOCRadar aims to make threat intelligence easier to use through automation, helping MSSPs protect smaller clients and improve their services.

Microsoft offers free cybersecurity boost to Europe

Microsoft is offering free cybersecurity help to European governments. This is in response to increased cyberattacks linked to actors from China, Iran, North Korea, and Russia. The program will improve intelligence-sharing on AI-based threats. It will also help prevent and stop attacks. Microsoft says AI can be used for both attacking and defending against cyber threats.

DeepMind CEO worries about AI misuse, not job loss

Demis Hassabis, CEO of Google's DeepMind, is more worried about AI falling into the wrong hands than AI causing job losses. He is concerned about the misuse of advanced AI and the lack of rules to control it. Hackers have already used AI to create fake voice messages. Hassabis believes there should be an international agreement on how to use AI safely. He also envisions AI assistants helping people with daily tasks.

Meta strengthens AI security with Llama Firewall

Meta created LlamaFirewall, an open-source tool, to protect AI agents from security threats. AI models face challenges like jailbreaks, prompt injections, and unsafe code. LlamaFirewall watches AI agents and stops threats in real-time. AI jailbreaks bypass safety measures, while prompt injections manipulate AI behavior. LlamaFirewall acts as a safety layer, analyzing AI inputs and outputs to prevent harmful actions.

Agentic AI safe adoption in networking and security

Agentic AI is growing fast and can help IT teams automate tasks and solve problems. These AI systems can use data and tools to achieve goals with little human help. They can fix vulnerabilities, resolve connectivity issues, and prevent problems. However, it's important to make sure AI doesn't cause unintended damage. Using digital twins, which are virtual copies of networks, can help verify AI behavior and ensure it works as intended.

Sources

AI Artificial Intelligence Agentic AI AI Security Cybersecurity AI Training AI Investment AI Regulation AI Governance AI Risks AI Threats AI Development AI Models Multimodal AI AI Agents Threat Detection Vulnerability Management Data Security LlamaFirewall Blackwell Chips Nvidia DeepMind Microsoft SOCRadar Zero Standing Privileges Continuous Monitoring Kill Switch SPIFFE Digital Twins Prompt Injection AI Misuse AI Safety AI Ethics