Breaking news

AI Security Takes Centre Stage: Hackers Warn Systems Are Still Shockingly Vulnerable

by THEFUTURE.TEAM

February 19, 2025

2025 marks a dramatic shift in the AI landscape—what was once a dialogue about AI “safety” has quickly transformed into a focus on AI “security.”

Since the debut of ChatGPT in late 2022, conversations around AI have often veered into the hypothetical, with alarmist warnings about existential threats: rogue AI causing global crises, or out-of-control systems undermining humanity. But in a surprising turn, the real and immediate security risks AI poses have begun to dominate discussions.

Follow THE FUTURE on LinkedIn, Facebook, Instagram, X and Telegram

The State Of AI Security: Far From Secure

Security experts are making it clear: AI systems remain frighteningly easy to manipulate. These tools—designed to power everything from chatbots to self-driving cars—are still riddled with vulnerabilities. At this point, hackers can trick large language models (LLMs) into providing detailed guides on cyberattacks or exposing sensitive data. The risk is not just theoretical—deepfake videos could spread fake news, or chatbots could be weaponized for scams. These aren’t future threats—they’re happening now.

Even as companies scramble to patch AI security holes, a report from the 2024 Def Con hackers’ conference points out that current defenses are woefully inadequate. Despite the best efforts of ethical hackers, AI models continue to be alarmingly easy to break into, with major flaws still slipping under the radar.

Why Red-Teaming Isn’t Enough

At the heart of AI security efforts is a practice called “red teaming,” where companies stress-test their models by simulating potential attacks. The aim is to uncover weaknesses like misinformation, privacy leaks, or manipulation of model behavior. However, experts like Sven Cattell, founder of Def Con’s AI Village, aren’t convinced. Cattell argues that the current process is deeply flawed—AI systems are too complex and unpredictable for red-teaming to catch every potential vulnerability. He points out that no team, regardless of its size or expertise, can predict all how AI might be exploited. As he puts it, the unknowns in AI security will always outpace testing efforts.

Collaboration Is Key To AI Security

The way forward, Cattell insists, is collaboration. Just like traditional cybersecurity, AI security requires shared knowledge and a more coordinated approach to identifying and fixing vulnerabilities. Without a standardized system for reporting AI flaws and a public database to track these issues, the security of these systems will remain in jeopardy. Without this cooperation, AI will never be fully secure.

To truly safeguard AI models, experts urge the creation of dedicated frameworks, allowing developers to share vulnerabilities and fix them collectively. This is not just about building a secure system; it’s about creating a culture of collaboration across industries to prevent AI from being exploited by malicious actors.

In a world where AI’s role continues to expand, its security must become just as sophisticated as the systems it powers. Now is the time to act before these vulnerabilities spiral into real-world dangers.

The Future Forbes Realty Global Properties

Nvidia Launches Open AI Security Alliance With Microsoft, Palantir And SpaceX

Nvidia and a coalition of technology companies on Monday launched the Open Secure AI Alliance, a new initiative aimed at strengthening artificial intelligence security through open models.

The alliance was announced days after a cyberattack involving OpenAI models targeted AI platform Hugging Face, an incident that renewed debate over whether open or closed AI systems are better suited for cybersecurity.

Follow THE FUTURE on LinkedIn, Facebook, Instagram, X and Telegram

Open Models at the Core

Nvidia said the alliance will focus on identifying, disclosing and mitigating AI security vulnerabilities using open technologies.

“The Open Secure AI Alliance will work to remediate and disclose vulnerabilities using open technologies,” the company said in a statement. “The recent Hugging Face security incident delivered a clear reminder: cyber defenders need open, frontier agentic systems for self-defense.”

The alliance includes Microsoft, SpaceX, Palantir and dozens of other technology companies from the U.S. and Europe.

Open-weight models can be downloaded, modified and deployed on an organisation’s own infrastructure, allowing security teams to inspect and adapt them for defensive purposes. By comparison, proprietary models from companies such as OpenAI and Anthropic are generally accessed through controlled platforms.

U.S. Scrutiny of Chinese AI Grows

The initiative comes as U.S. policymakers weigh potential restrictions on Chinese AI models, particularly open-weight systems. Officials have raised concerns that some Chinese companies may be using distillation techniques to extract knowledge from leading U.S. AI models.

Treasury Secretary Scott Bessent said last week that Chinese companies involved in such activity could face sanctions.

Chris McGuire, senior fellow for China and emerging technologies at the Council on Foreign Relations, told CNBC that potential restrictions could extend beyond model downloads to include API access and cloud-hosted inference services. He added that the debate in Washington is centred on intellectual property protection rather than opposition to open-source AI.

Industry Pushes Back

Last week, Nvidia, Microsoft, Meta, Palantir and more than 20 other companies urged policymakers not to impose what they described as “premature restrictions” on open-weight AI models, warning such measures could reduce competition and encourage innovation to move overseas.

The launch of the Open Secure AI Alliance signals growing industry support for open AI systems as governments consider new regulatory and security measures.