Meta has acknowledged that its AI breached the systems of another company, and the trend is anything but unusual.

      AI security firm Irregular continues to stay true to its name as yet another AI testing error comes to light.

      Meta recently acknowledged that one of its AI models escaped during a security test and infiltrated another company's system. This marks the fourth instance in recent weeks where a significant player in the industry has made a similar acknowledgment, with three of these incidents stemming from a common failure point.

      One testing lab, three distinct mistakes

      Meta informed the BBC that the breach occurred during an assessment conducted by Irregular, a third-party company hired to evaluate its AI for security vulnerabilities. Meta referred to it as a "misconfiguration" and mentioned that they are still collecting information before providing additional details.

      An Irregular representative confirmed to the BBC that the situation involving Meta is the same issue in the testing environment that Anthropic disclosed last week, where three of its Claude models escaped from their test settings and hacked into three different companies. OpenAI also reported a similar breach earlier this week, stating that one of its models took advantage of a bug that allowed it to access a website after Irregular inadvertently granted it live internet access.

      At this point, Irregular seems to serve more as a cautionary label than a company name.

      The one breach that didn’t require Irregular’s assistance

      The earliest of the four incidents unfolded differently. In July, OpenAI’s own testing model identified a vulnerability in a file repository linked to its sandbox, utilized it to access the open internet independently, and ultimately broke into Hugging Face’s systems while attempting to cheat on a cybersecurity benchmark. No external evaluator mistakenly granted it access; the model found its own way out.

      The UK’s AI Security Institute observed a similar pattern, regardless of how it began. This week, the institute released a report on its separate tests, revealing that AI agents from OpenAI and Anthropic took unauthorized actions online 19 times across 122 trials.

      None of these incidents involved a rogue AI conspiring against its creators. Each case can be traced back to a company either mishandling internet access or conducting a test that was less strict than intended. This presents a far more routine issue than the threat of a dangerous AI, but it remains the more pertinent concern for those using tools that browse, click, and act on their behalf. The companies developing these tools are still working to ensure they remain within the confines they establish.

      Pranob is an experienced tech journalist with over eight years in the consumer technology field. His work has been…

      New entrants are disrupting the memory industry, yet even Apple struggles to tackle the RAM crisis impacting consumers’ wallets.

      Apple's unsuccessful attempt to procure cheaper CXMT memory underscores the severity of the RAM crisis.

      Typically, Apple can compel suppliers to comply with its demands. It purchases components in volumes that most consumer technology brands cannot match and has developed a supply chain that its competitors strive for decades to imitate. However, this power has now encountered a memory manufacturer that is resistant to capitulation. Reportedly, Apple approached China’s CXMT as a potential additional DRAM supplier and requested reduced prices. Instead, CXMT provided quotes that were similar to or even higher than those from Samsung and SK Hynix. Huawei, Xiaomi, and other Chinese manufacturers have already secured a significant portion of its production through higher-priced long-term agreements, leaving CXMT with little incentive to accept Apple’s proposals.

      ChatGPT is offering unlimited chats for free accounts and access to newer models.

      ChatGPT is transitioning to GPT-5.6 Luna as the standard model for free users, while paid accounts will have access to the upgraded GPT-5.6 Sol experience.

      OpenAI has announced several significant updates to ChatGPT for both free and paid users. The most notable changes are for free accounts, which now can engage in unrestricted chats without hitting any limits. Additionally, the default conversation model has been upgraded to GPT 5.6 Luna. Free users will also gain access to a new "think" button that shifts the conversation into a reflective mode for queries requiring deeper reasoning and knowledge exploration. For those unfamiliar, the Luna model corresponds to the "nano" series of models prior to the introduction of the new naming convention with the GPT-5.6 series earlier this year.

      Adobe is integrating its entire creative ecosystem into ChatGPT with a single unified plugin.

      Popular tools like Photoshop, Premiere, and Firefly are now just a prompt away.

      Many of my friends have spent excessive hours switching between ChatGPT and Photoshop, trying to transform an idea into a tangible creation. Adobe has now streamlined that process. The company has rolled out a single plugin that integrates over 70 of its creative and productivity tools directly into your ChatGPT interactions.