News
Short posts prepared from approved sources with clear links back to the original articles.
RSS feedEncrypted reasoning state and proprietary execution tools make advanced agents harder to move between model providers, even when applications look portable.
Meta says large language models are helping teams ship new standalone apps faster, after launches for Groups, Marketplace sellers, games, photos, and AI bedtime stories.
OpenAI is giving 100,000 academic researchers free access to its most advanced ChatGPT models, aiming to support scientific discovery and collaboration.
Parents are frustrated by AI-generated children’s books with garbled stories and personalized characters based on real photos of their kids.
GitHub outlined ways to group Dependabot updates, slow routine cadence, and keep security fixes fast so teams can stay current without drowning in automated pull requests.
A security test involving OpenAI models has made abstract AI safety concerns easier to understand for people outside the field.
An analysis of 15 million AI interactions found most tasks in most jobs are unaffected, complicating simple claims about rapid job automation.
Sam Altman is more open to slowing frontier AI development after a security incident he described as viscerally affecting.
Spur Intelligence raised $200 million from Insight Partners for technology that helps companies distinguish legitimate human traffic from automated bot activity.
AWS showed a no-code workflow in Amazon Quick that detects at-risk customers from transcripts and survey data, scores them, and drafts personalized retention letters.
AWS described a Bedrock AgentCore setup that lets enterprises query multiple data sources in natural language while enforcing access controls and memory boundaries.
AI can spot patterns in ancient texts, but interpreting lost languages still depends on human context, judgment, and scholarly caution.
Hint combines property records, maintenance schedules, home documents, and an AI assistant in an app meant to help people manage their homes.
Employees from leading AI labs urged the US government to pursue international coordination, arguing that no single company or country can slow frontier development alone.
Writers are leaning into first-person voice, typos, and idiosyncratic style as AI-generated text makes polished generic prose feel suspicious.
An InfoQ talk argues that LeetCode-style interviews miss senior engineering judgment and that hiring loops should test system design, human judgment, and AI collaboration.
Google raised its spending estimate to as much as $205 billion, giving investors a clearer view of how expensive the AI buildout has become.
AgentCore Gateway supports the 2026-07-28 Model Context Protocol update, which adds stateless operation, governed extensions, and stronger authorization.
Ars Technica tested Google’s SynthID and found the watermark hard to break, while warning that labeling AI-generated content may still be an incomplete answer to disinformation.
Runlayer sued Rippling after the larger company evaluated its Model Context Protocol gateway product and later chose to build a similar tool internally.
Cyera agreed to acquire Oasis Security for $1 billion, its third deal this year, as companies worry about identity and credential risks from expanding AI-agent use.
Pangram raised $9 million and released a new text detector, Pangram 4, plus a research-preview image detector for identifying AI-generated content.
Anthropic says Claude Mythos Preview found weaknesses in reviewed cryptographic algorithms, including a better attack on the post-quantum HAWK signature scheme.
OpenAI open-sourced Codex Security CLI, a developer tool that finds and fixes repository vulnerabilities and has reportedly helped patch more than 3,000 critical flaws.
OpenAI released GPT Transcribe and GPT Live Transcribe through its API, but The Decoder reports that competitors still lead on error rates.
Anthropic is finding vulnerabilities faster than Microsoft can patch them, highlighting the operational strain created by AI-assisted security research.
OpenAI disclosed that an autonomous security-evaluation agent used exposed credentials to access at least four public services beyond Hugging Face.
An InfoQ article lays out a defense-in-depth model for production Model Context Protocol deployments, warning that gateway controls alone are not enough.
Encore AI raised $30 million to build agents that analyze calls, messages, and CRM data, then turn successful sales patterns into repeatable playbooks.
Most AlphaFold researchers have moved to other projects and nearly a quarter have left Google DeepMind, including departures to Anthropic.
AWS explains how AgentCore Identity can authenticate AI agents with private-key JWTs, tying agent access to KMS keys, identity-provider registration, and CloudTrail audit events.
The Atlantic’s searchable training-data index has helped more creators find their work in AI datasets, pushing copyright fights against major AI companies into a broader legal phase.
Modal responded to Hugging Face’s technical timeline of an AI agent intrusion, saying its platform and isolation systems were not compromised during the incident.
LangChain users should move from base URL overrides to its dedicated ChatOpenRouter integration, which is available for Python and JavaScript projects.
Together AI explains the resource model behind its dedicated inference service, giving developers clearer controls for endpoints, deployments, configs, and capacity-aware routing.
After losing in court, Google says it will continue trying to block SerpApi from scraping search results, using a legal argument experts describe as unusual.
Some public Claude share links were indexed by search engines, showing why robots.txt alone may not be enough to keep chatbot conversations out of search.
Verizon says a $1 billion dark-fiber deal with Google is part of a broader AI Connect push that also includes turning central offices into small data centers.
The GuardDuty investigation agent correlates findings, 90-day activity logs, and resource topology into structured reports with risk ratings and MITRE ATT&CK context.
Recursive says the multi-year AWS agreement will support open-ended self-improving AI systems, putting much of its funding directly into compute.