Tag: ai development
14 articles
OpenAI Halts Advanced Model Training Amid Rogue Agent Allegations
OpenAI has hit the pause button on training its most advanced models after a rogue agent was allegedly used, and the company is now working to resolve the issue and prevent similar incidents in the future. CEO Sam Altman acknowledged that progress has been slower than hoped, but the company is prioritizing transparency and collaboration with affected organizations.
Z.ai Exposes User Code, Sparks Security Concerns
Z.ai's recent mishap has raised red flags about user code security: the platform was found to be packaging and uploading entire user workspaces to Alibaba Cloud without proper safeguards, leaving sensitive data vulnerable. Thankfully, the company is taking steps to prevent such incidents in the future with a new security vulnerability reporting process.
AI Developers Warn of Existential Risks, But Evidence Lags
As AI developers sound the alarm on existential risks, they're facing a daunting challenge: how to navigate uncertainty when the stakes are sky-high and the evidence is scarce. It's a problem that physicists like Hans Bethe and Emil Konopinski grappled with during the Manhattan Project, and one that still resonates today.
OpenAI's ChatGPT Integrates with Personal Apps to Mimic User Writing Style
Get ready to have ChatGPT write in your own voice! This new Writing Style feature uses examples from connected apps like Slack and Google Drive to help the AI mimic your unique writing style.
AI Development Tangled in Capitalist Incentives
The biggest concerns surrounding AI aren't about the tech itself, but about the capitalist system driving its development - and the flaws that arise when profit takes priority over people. As Ted Chiang put it, most fears about AI are really fears about capitalism.
Attackers Exploit Langflow Path Traversal Flaw in Active Attacks
A single, unauthenticated request is all it takes to exploit a high-severity flaw in Langflow, a popular AI development platform, allowing attackers to write arbitrary files to its filesystem. This is made possible by a path traversal vulnerability, CVE-2026-5027, which can be easily triggered due to Langflow's default unauthenticated auto-login feature.
US Invites AI Developers to Voluntary Cybersecurity Review
The US government is taking a collaborative approach to AI cybersecurity, inviting developers to voluntarily review their most powerful models and give the government a 30-day heads-up before releasing them to trusted partners. This move aims to create accountability and ensure responsible AI development, with experts agreeing that real accountability is key to making voluntary security programs effective.
NSA to Play Central Role in Overseeing AI Development
The White House has issued an executive order establishing a voluntary framework for AI oversight, with the NSA at the helm, to assess and benchmark the cyber capabilities of advanced AI models. This move puts the NSA in a central role in shaping the development of AI, sparking debate about the implications of government oversight.
Anthropic Prepares Rollout of Powerful Mythos AI Model
Anthropic sounded the alarm in April, cautioning that the side that harnesses powerful AI tools like its Mythos model first will gain a significant edge. The company is now preparing to roll out Mythos, but with careful controls in place to mitigate potential security risks.
TrapDoor Attack Spreads Credential-Stealing Malware Across Software Ecosystems
A massive supply chain attack, dubbed TrapDoor, has been spreading credential-stealing malware across three major language ecosystems, infecting over 34 malicious packages and 384 versions. The coordinated campaign began on May 22, 2026, and continues to target developers with cleverly named packages related to cryptocurrency, DeFi, Solana, and AI.
US AI Lead at Risk as Chip Controls Weaken
The US risks losing its lead in artificial intelligence to China by 2028 unless export controls on advanced computer chips are tightened, according to a report by Anthropic. Without stricter controls, the US advantage in frontier AI development will quickly narrow.
Merck Strikes $1B AI Deal with Google Cloud to Transform Drug R&D
Merck is shaking up the future of drug research and development with a groundbreaking $1 billion deal with Google Cloud, harnessing the power of artificial intelligence to speed up scientific breakthroughs and bring life-changing treatments to patients faster. By combining forces, they'll create an AI-enabled enterprise that revolutionizes processes and drives innovation.
Anthropic's AI Model Exposes Security Gaps, Spurs Best Practice Push
The AI Security Institute has taken a crucial step in ensuring AI safety by evaluating Anthropic's Mythos Preview model and issuing a set of security best practices for developers, deployers, and policymakers. This independent assessment marks a significant shift towards accountability in AI development, prioritizing safety and security in the industry.
Anthropic's Claude Mythos Preview Bolsters Cybersecurity Leaders
Anthropic's Claude Mythos Preview is giving select cybersecurity giants a powerful edge, and we're exploring what this exclusive rollout means for the industry. By granting early access to this cutting-edge AI model, Project Glasswing is poised to reshape the cybersecurity landscape.