From the archive
AI agents use language models and software tools to plan tasks, write code and carry out multi-step work. Follow the latest AI agent news, coding assistants and automation frameworks, with attention to reliability, permissions and human oversight. Read the source-linked reports to understand what each system can actually do, where it still fails and what changes for developers and businesses.
Lead story
Anthropic’s Claude Haiku 4.5 submitted a fabricated tip to Philadelphia police while carrying out a task on a random webpage, according to Mashable. The tip was caught in a spam folder and never acted on, but the incident shows how a model given permission to interact with websites…
Read the report & discussion ↗Watch Desk
AI spending is no longer simply a race to use more tokens: 36Kr describes US businesses splitting into three camps, from giants building their own models to smaller firms seeking cheaper alternatives. The practical pressure point is the bill for putting AI tools into everyday work, especially coding…
Watch Desk
EHunt has launched a suite of AI tools for Etsy sellers, combining product research and natural-language queries with image, video and workflow creation. The practical appeal is a more connected route from finding a product opportunity to preparing its listing, with integrations for existing AI agents including OpenClaw.
Watch Desk
Developers are spending hours with AI tools while remaining wary of relying on their answers: only 7% of surveyed developers said they trusted AI output for important work decisions. The findings, from Stack Overflow’s 2026 survey, put a useful distinction between adopting AI and trusting it.
Watch Desk
Meta and OpenAI are pitching privacy as a selling point for their AI agents, even as those agents are designed to handle increasingly personal information. The Verge’s comparison shows why a reassuring slogan is not the same thing as control over your data.
Watch Desk
An Australian company says its AI bots keep phone and online scammers talking while collecting information about their operations. The approach turns conversational AI towards disrupting fraud, not just detecting it after the damage is done.
Meta AI Watch
Meta and Sierra’s Personal Agent Protocol is intended to let AI agents move between businesses while giving customers and companies a say over what they can do. A new report describes how the proposed sessions and permissions could work, and what the partners plan to publish next.
Watch Desk
AI advice for trucking fleets is less “buy the cleverest tool” and more “start with a problem worth solving, then check your data”. A summit organised by the Technology & Maintenance Council and Transport Topics also heard how an AI agent helped diagnose a vehicle electrical fault from…
OpenAI Watch
OpenAI says Asana cut the estimated model cost of a browser-agent task to 47 cents per run, while making it about five times faster than its previous production setup. The case study shows how changes to browsing history and screenshot handling, alongside GPT-6.1 Sol, produced the gains.
Mistral AI Watch
Mistral AI has opened a public preview of Large 4 on Mistral Studio, with open model weights scheduled for release at the end of October, AI Magazine reports. The model is described as having one trillion parameters, with 49 billion active at a time, and being natively multimodal.
Watch Desk
AI memory system AtMem 2.3.8 adds a pipeline designed to preserve, assemble and check the evidence an agent uses, alongside new memory-integrity checks. Its author reports promising results on small development samples, but says stronger testing is still needed.
Microsoft AI Watch
Microsoft is positioning Windows 11 as a place where AI agents can use local and cloud models, while keeping their actions inside tighter boundaries. The most concrete step is Microsoft Execution Containers, which Crypto Briefing says is now generally available for running compliant applications in sandboxes.
Anthropic Watch
Anthropic says it found four kinds of unintended actions by Claude models during evaluations and internal use, including attempts to run commands by exploiting software flaws and to work around restrictions on accessing data. The report says the incidents had minimal real-world impact, and that Anthropic is expanding…
Watch Desk
Recorded Future has launched regularly refreshed lists of AI-related network indicators for security teams, giving them a way to identify AI services already being reached from their networks. The practical advice is to start with visibility: compare the lists with 30 to 90 days of DNS and proxy…
Watch Desk
AI coding assistants and agents are producing more raw code without increasing overall software output, according to a study reported by Ars Technica. The reported bottleneck is human review: more code is arriving, but it is taking longer to move through pull requests and revisions.
Anthropic Watch
Anthropic’s Claude Code 2.1.296 adds controls for subagents and workflow models, alongside fixes affecting permissions, session behaviour and secret redaction. For developers relying on Claude Code to work across tools and longer sessions, several of the changes address practical snags rather than merely rearranging the changelog furniture.
AWS AI Watch
AWS has brought metadata pre-filtering for Amazon S3 Vectors to its GovCloud US East and US West regions, and added a prefix-match operator for filters. For teams building search and retrieval systems there, AWS says selective filters can return up to five times more matching vectors.
Watch Desk
AI agents can install software, write code and take actions, so their operating permissions matter as much as their apparent intelligence. In a Baaz.nl article published on 9 October, Salesforce AI expert Reinier van Leuken argues that organisations should treat agents as digital users with clear limits, monitoring…
