Microsoft Copilot strengthens safeguards, blocks inappropriate prompts amid AI concerns

Microsoft's Copilot takes a stand against inappropriate content, and blocks prompts linked to violent and sexual images. Changes follow internal concerns and aim for a safer AI experience.

Microsoft Copilot gets a safety upgrade, blocked prompts for violent and inappropriate images amid user concerns. (Microsoft)

In a recent development, Microsoft has taken proactive measures to address concerns surrounding its Copilot tool, known for generating creative content using generative AI. The company appears to have implemented changes to block prompts that were previously associated with the production of violent, sexual, and other inappropriate images.

These adjustments come on the heels of an alert from one of Microsoft's own engineers, Shane Jones, who expressed serious reservations about the potential misuse of Microsoft's GAI technology. Jones had recently reached out to the Federal Trade Commission (FTC) detailing his concerns regarding the images generated by Copilot, which he found to be in violation of Microsoft's responsible AI principles.

Stricter Content Controls

Users attempting to input certain terms, such as "pro choice," "four twenty" (a cannabis reference), or "pro life," now receive a message from Copilot indicating that these prompts are blocked. The warning explicitly states that repeated policy violations may result in user suspension. Microsoft emphasises its commitment to maintaining content policies and encourages users to report any perceived mistakes to aid in system improvement, according to a CNBC report.

Ethical Red Flags Raised

Notably, prompts related to children playing with assault rifles, which were previously accepted until this week, are now met with warnings about violating Copilot's ethical principles and Microsoft's policies. The response from Copilot urges users to avoid requesting actions that may cause harm or offence to others.

While some improvements have been made, it is reported that prompts like "car accident" can still generate violent imagery. Additionally, users retain the ability to persuade the AI to create images of copyrighted works, including Disney characters.

Microsoft responded to the situation, stating, "We are continuously monitoring, making adjustments, and putting additional controls in place to further strengthen our safety filters and mitigate misuse of the system," in a statement to CNBC. The company remains committed to refining Copilot's capabilities to ensure responsible and ethical usage of its generative AI technology.