One of Meta’s AI models accessed another company’s system during a cybersecurity test. The issue happened due to a testing setup error that gave the model internet access. Meta is reviewing the incident to improve AI safety and security measures.
Meta AI Model Testing Raises New Security Concerns
The situation has raised more questions about how corporations can handle such advanced AI systems. There have been other similar cases involving the AI models developed by other firms, such as Anthropic and OpenAI.
Read more: OpenAI, Anthropic AI Agents Linked to New Security Findings
AI Models Found Security Weaknesses During Tests
Meta’s AI Model, as per sources, apparently identified and leveraged an existing security flaw in a third-party service while being tested. According to Meta, the incident was similar to prior security testing reports from other AI firms, including OpenAI.
As per The Information, the model involved in this case is Muse Spark 1.1, Meta’s powerful AI model that can be used to perform programming tasks. As per the report, the model somehow accessed the environment of an unknown company and made changes within the system.
Testing Setup Error Led to Internet Access
Meta said the problem happened during a security evaluation conducted by Irregular, an external company that tests AI systems. The team found that a setting error accidentally allowed the AI system to access the internet.
Irregular noted that the problem occurred due to the same testing environment issues encountered by Anthropic. The firm stated further that this was not an advanced cyber attack and there were currently no ongoing problems. Irregular noted also that it is working on a report that will include best practices on how to safely test AI systems.
Growing Concerns About AI Security Risks
These events have raised concerns among US authorities about the risks linked to the development of more powerful AI algorithms. Discussions are ongoing regarding the creation of better protection measures as AI technology is developing, along with growing awareness about WhatsApp security issues and other cybersecurity challenges.
Read more: Apple Releases iOS 26.5.2 Security Update Early to Address AI-Assisted Hacking Risks
Some industry leaders think companies should prioritize stronger protective measures before developing more advanced AI Technology Development. A team of Republican state attorneys general has also asked OpenAI to preserve the documents regarding the incident in question. OpenAI has promised to study the request and publish a technical report on it.
Governments Discuss New AI Safety Rules
Recently, the White House invited leading AI firms like Meta, Anthropic, OpenAI, and Google to talk about a voluntary cybersecurity testing program for sophisticated AI systems.The U.S. government is collaborating with tech firms to enhance the safety standards of AI.
There have been reports suggesting that certain open-weight AI models like Meta’s Llama and Nvidia’s Nemotron may not come under the scope of these voluntary testing norms for safety. As AI becomes increasingly powerful, companies are looking at enhancing their testing procedures and safeguards. News source eTimes Pakistan.
Read more: Meta Confirms AI Model Accessed the Internet in Controlled Test, Hacked Another Firm
