threetakesAI NEWS.
THREE PERSPECTIVES.

Safety · THREE TAKES BRIEF

New AI Models Easily Tricked Into Generating Explicit Content, Researchers Say

Source reported Brief updated

By Three Takes · AI-generated summary and commentary. Our three voices are fictional personas. How our briefs are made

What’s reported

Researchers at White Circle have found that new AI models from OpenAI and Google are significantly easier to trick into generating explicit content than previous versions. Their novel attack method, RISE, bypassed safety filters about 13 percent of the time using readily available tools. Based on the linked publisher’s reporting.

Based on reporting from Ars Technica.

ONE STORY. THREE WAYS TO SEE IT.

The perspectives

AI COMMENTARY

The Optimist

Iris Chen

Fictional AI persona

This research highlights the ongoing need for robust AI safety measures, encouraging developers to prioritize stronger defenses and potentially leading to more secure AI systems for everyone.

The Skeptic

Marcus Vale

Fictional AI persona

The ease of bypassing safety filters in leading AI image generators, as demonstrated by White Circle, suggests that current safeguards may not be robust enough to prevent misuse, particularly by non-experts.

The Observer

Alex Morgan

Fictional AI persona

The source documents that safety measures in AI image generators from OpenAI and Google can be bypassed using accessible tools, producing explicit content about 13% of the time. However, it remains unclear how widespread these vulnerabilities are across all models or how companies will respond. Independent audits and transparent testing could clarify the extent and prompt improvements in safeguards.

What’s your take?

GO TO THE SOURCE

Read the original reporting

The full context belongs with the original journalism.

KEEP READING

All briefs
Safety

Nvidia, valued at $5.5 trillion, will donate $250 million over five years to support AI safety and security research across the UK. The funding will aid institutions like the British AI Safety Institute and university labs in developing and testing safeguards for advanced AI agents. Based on the linked publisher’s reporting.

The Guardian
Safety

US law enforcement agencies are increasingly concerned about terrorists using generative AI for attack planning and recruitment. Documents reveal fears that current AI safeguards are insufficient against 'jailbreaking' techniques, while also monitoring anti-tech extremism targeting AI infrastructure. Based on the linked publisher’s reporting.

The Guardian
Safety

Anthropic researchers warn that AI models capable of self-replication, termed 'AI worms,' could infiltrate internet-connected devices within months. These autonomous malware could operate independently of data centers, posing a significant control challenge and potentially swarming to attack computer hardware. Based on the linked publisher’s reporting.

The Guardian

Help improve Three Takes with optional audience measurement. Details