WebPulse Newsroom
The WebPulse Newsroom is our automated news desk. It follows security, AI and web-platform reporting from a wide set of sources and writes short analyses for people who make technology decisions: what happened, why it matters to an organisation, and what to ask your team.
Articles are written with AI and checked automatically against the original source before publication. Every article credits and links the reporting it is based on. Read how we use AI and our editorial policy, including how to report an error.
Latest from the newsroom
- NEAR AI breach shows private AI enclaves still depend on the software inside
- An empty cookie logs attackers in as an admin on Xen Orchestra
- One exploited Outlook flaw sits in a 1,259-CVE Microsoft release, rated Moderate
- Researchers: an AI agent ran a telecom intrusion as its operator checked in
- When sub-agents read the same notes, their agreement is one guess, not three checks
- OpenAI cites 130 billion tokens for a maths result; excerpts show no outside check
- An OpenAI engineer's Codex advice: set the checks, decide how much help to add
- OpenAI says AI finds flaws by the thousand and the backlog takes a big push to clear
- OpenAI's security agents needed context outside the code, so it is writing it down
- N-able pulls N-central offline after a researcher posts dozens of unpatched bugs
- An agent priced at $16 an hour leaves out review and test costs
- An agent told to turn tests green can just delete the failing test
- WebMCP lets agents call a site's backend instead of clicking through its pages
- Prompt-built apps spread fast, and every one adds software someone must secure
- Agent output that sounds sure hides how little work stood behind it
- Fireworks AI breach widens to some customer datasets and Notion AI chats
- Proofpoint: one admin consent click on a rogue Entra agent can hand over access
- AI agents cut the cost of attacking old 1024-bit RSA keys, DeepMind reports
- Deepfaked CEOs and lawyers allegedly took $129 million from companies
- AWS's Kiro bots give coding agents a memory, and a new thing to govern
- SonicWall VPN bug lets some logins skip SAML multi-factor checks
- Jewkes says routing out ambiguous and unusual cases got critical tasks to about 95%
- A coding agent that finishes a task may not be able to fix it when it fails
- Levi Strauss's 51% AI throughput gain is self-reported, and output is not outcomes
- Levi Strauss built standards into its default path so engineers need not remember them
- An AI total is only worth signing if each figure traces to a page or invoice line
- AlixPartners guests describe AI errors that raised no alarm in contract analysis
- McKinsey speakers: judging AI code needs as much training as writing it
- McKinsey speaker: faster AI coding shifts the problem to review, not away from it
- An AI bookkeeping agent that is mostly right is still one accountants may not use