AI access is spreading—so are the questions about trust and control

Months after launch, Penn’s secure AI portal has gained features and reports usage, but many students and faculty still don’t know much about it. For institutions building AI tools, availability is only the first step; people also need a reason to try them and confidence in how they work.

Kong’s new Token Vault is designed to reduce scattered credentials and limit the damage if an AI agent is compromised. Builders connecting agents to enterprise systems will need to treat secret management as part of the product, not an afterthought.

Leverate has introduced a CRM for CFD brokers with an AI assistant. It’s another sign of AI moving into specialized business software, where usefulness will depend on how well the assistant fits the work brokers already do.

LG AI Research has made the API for its upgraded EXAONE 2.5 available free of charge. Lower-cost access can make it easier for developers to test a model, though teams will still want to assess its fit for their own tasks.

Muse confirmed that data from its sleep and meditation headband helped train an AI model of the brain, based on about 50,000 records. For users and product teams alike, the story is a reminder to ask clearly what sensitive data may be used for.

Globe Telecom is offering millions of mobile and broadband customers free access to AI tools during the holiday season. Bundling could bring more people into contact with these tools, while making the range of available services and their terms worth checking.

Perplexity’s Teammates feature is currently limited to individual Max subscribers, with Pro and enterprise users waiting for wider access. That staged rollout gives early users a head start, but leaves teams considering agent workflows wondering when they can evaluate it.

ETRI has made its TANGO development tool open source, aiming to let industry workers build AI models without specialized expertise. Open access may help more people experiment, though teams will still need to decide how to validate what they build.

IIT Delhi has launched a six-month certificate programme covering cybersecurity and AI security. As AI systems become part of more workflows, training that connects security fundamentals with AI-specific risks could be valuable for practitioners.

Google Cloud is adding identity, perimeter, and detection controls intended to help organizations secure AI-driven systems and respond to issues faster. Alongside tools for managing credentials, these controls point to the growing need to secure agents across the systems they can reach.