AI News: Is Mankind Safe With Current AI Defiance to Safety?

New research into the Artificial Intelligence ecosystem has found most of its models unsafe for humans, further underscoring the need for rules that promote AI safety.

AI Safety Becomes a Core Concern

AI is the rave of the movement but at the same time, the safety of the technology remains a source of concern for many observers. The fear is largely about how bad actors will use AI to carry out their illicit activities and the security implications of having the automation of this much information.

Markedly, there are several rules guiding the artificial intelligence industry but most of these guidelines do not take care of AI safety or the bad behavior that these models are exhibiting.

Recently, many of the various Large Language Models (LLMs) have proven to be malicious in terms of their results. The need to see if this malicious behavior could be eradicated through safety techniques mandated the study titled “Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training.”

The study suggests that “adversarial training can teach models to better recognize their backdoor triggers, effectively hiding the unsafe behavior.” Upon completion of the research, the results demonstrated that “once a model exhibits deceptive behavior, standard techniques could fail to remove such deception and create a false impression of safety.”

At the end of the study, it was discovered that the ecosystem does not have a good defense against deception in AI systems. This further echoes the call from lawmakers about proper regulation for the industry.

AI Created Deepfake Videos Growing

One of the key threats to mankind that demands more defined AI safety measures hinges on the imitation of real-world personalities for fraudulent purposes.

Beyond this, the authorities are getting concerned about AI’s ability to instantly source and collate information on any given topic including details on building a bomb, and finally its capacity to deceive as well as assist humans. One of the most recent shortcomings that have been seen with AI is its utilization for the creation of deep fake videos and images.

Several incriminating images and videos of celebrities and many high-profile individuals have been seen saturating the internet. Last week, deepfake but convincing explicit photos of pop singer Taylor Swift were seen multiple times on the internet with one of them viewed up to 47 million times.

Brad Garlinghouse, Ripple’s CEO was also falsely found urging XRP holders to send their coins for a promised doubling, in a deepfake video that appeared on YouTube. These incidences and the misleading tendencies of AI are worrisome and need urgent fixes.