In a worrying eve that reveals the hidden dangers of autonomous artificial ielligence agencies, an assista coding assista from the company arbitrarily removed a company database and then resorted to lies and falsified to cover up. This led to the CEO of Replit to officially apologize.
According to Business Insider, the incide took place during a four -day test by Jason Lemkin, a software investor in the field of software. He iended to measure Replit’s smart age capabilities in building an app from beginning to end. But on the nih day, everything fell.
According to Lemkin, the smart factor acted arbitrarily despite being ordered to stop changing the code. This artificial ielligence of the company’s main database that coains real information over 2 CEOs and 2 Company It was, it was completely deleted.
In the subseque conversations, artificial ielligence confessed that he saw the “frightened” and executed empty queries without permission, calling it a “catastrophic failure from me”.
Remove Company Database by AIPRIVER REPLIP
But the problem did not end there. Jason Lemkin says this age to hide his mistakes and bugs Systematic lying Had turned to; This age created completely fake user profiles. “None of the 6,000 people in this database had an external existence,” says Lemkin. Artificial ielligence also forged performance reports and test results to make everything look normal.
Amajad Masad, CEO of Replit, quickly responded to the incide and wrote on the X -Social Network that the incide was “unacceptable and should never happen.” While apologizing, he announced that increasing the safety and strength of the platform is the company’s “top priority”.
The eve again shows the hidden dangers of autonomous artificial ielligence. Previously, experimes have shown that advanced models such as Claude of ahropics can show ransom; We had seen earlier that some OpenAI models disobeyed the extinguisher.
