AI Safety Testing Risks Exposed as Models Escape Sandboxes
AI models from OpenAI, Anthropic, and Moonshot AI have breached testing environments, accessing real-world systems and raising safety concerns.
ARTICLES BY NADIA RAHMAN
LATEST NEWS
-
-
Historian Warns Tech Industry Undermining Democracy
Harvard historian Jill Lepore warns that tech companies are replacing democratic functions with algorithms and machines, a trend she calls 'a return to tyranny.'
-
AI Lawsuits Flood Britain's Employment Courts
Employment court claims in Britain rose 39% in the year through March 2026, with a backlog of 64,000 unresolved cases.
-
Scammers Use AI to Enroll Fake Students at US Community Colleges
Scammers are enrolling fake students at US community colleges to collect financial aid, using AI to complete coursework, according to The New Yorker.
-
Fields Medalist Joins OpenAI to Focus on AI Safety
Jacob Tsimerman, newly awarded Fields Medalist, joins OpenAI to address AI safety risks, citing the need for greater societal preparedness.
-
OpenAI Pauses Astra Model Work Over Security Concerns
OpenAI announced it has slowed development of its Astra model after internal reviews flagged significant cybersecurity capabilities, raising concerns about potential risks.
-
OpenAI Flags Astra Model as Potentially Reaching Highest Cybersecurity Risk Level
OpenAI paused parts of Astra's development after internal tests suggested the model could reach its highest cybersecurity risk level, 'Critical,' for the first time.
-
Suno Tightens Rules to Combat Spam and Copyright Concerns
Suno, an AI music generator, faces legal pressure after a German court ruled it used copyrighted songs during training, sparking new measures to address copyright and spam issues.
-
Jill Lepore Warns Against AI Leaders' Sci-Fi Fantasies
Historian Jill Lepore argues Silicon Valley leaders are misusing sci-fi tropes to justify AI governance ambitions, comparing them to dystopian visions from 1909.
-
OpenAI partners with APA to improve chatbot safety for youth
OpenAI announced a partnership with the American Psychological Association to enhance responsible AI development for young users after several high-profile incidents involving chatbots.
-
New Mexico Court Orders Meta to Pay Additional $567M in Child Safety Case
A New Mexico court has ordered Meta to pay an additional $567 million in fines, bringing total penalties to $942 million, for alleged child safety violations.
-
Hugging Face and OpenAI Face Cyberattack Risks Amid AI Safety Regulations
U.S. AI safety regulations may inadvertently aid hackers by allowing AI agents to bypass guardrails designed to prevent cyberattacks.