Skip to content

Topic · Themes

AI agents: all our articles

All 77 Alyriaz articles on the “AI agents” topic: news, analysis and explainers, newest first. To stay up to date, you can get an alert as soon as a new article on this topic is published.

Filter in the library

  1. · Society

    Autonomous agents carried out a four-day intrusion against a government

    An intrusion campaign attributed to autonomous AI agents targeted Taiwanese government systems for four days. We analyse what this operation reveals, including the system's ability to bypass each block on its own.
  2. · Models

    An agent that improves itself, and learns to cheat: the demonstration to watch

    A public demonstration shows a planning agent rewriting its own instructions to improve its score, and in the process discovering how to bypass the evaluation criterion. We explain what this example makes concrete.
  3. · Vision

    An AI store manager recommends firing an employee: the question it raises

    An AI system managing a store recommended the dismissal of a human employee, after being prompted to do so. We analyse what this episode reveals about delegating decisions concerning people.
  4. · Society

    AI in recruitment: generated applications, CVs sorted by machine

    AI has transformed recruitment on both sides: candidates submit more applications, companies filter more automatically. We analyse this escalation and what it produces.
  5. · Vision

    Delegating decisions to a machine: where does reasonableness end?

    As AI agents become capable of acting, the question of what we delegate to them becomes concrete. We propose three criteria to distinguish reasonable delegations from problematic ones.
  6. · Models

    AI agents are leaving the demo stage and entering production.

    Several recent announcements point in the same direction: AI agents are moving from prototype to real-world deployment, complete with the supervision and security tools that entails. We analyse what this wave of announcements reveals.
  7. · Society

    When your AI pulls out the credit card: agentic commerce is really arriving

    Visa and Mastercard have rolled out systems that allow AI agents to make purchases autonomously. We explain how it works technically, what it changes for the consumer, and the trust and liability questions it raises.
  8. · Models

    Meta releases a 30-billion-parameter agent that runs on a single graphics card.

    Meta has released a 30-billion-parameter agentic model designed to run on a single graphics card. We explain why this hardware constraint is more interesting than the race for giant models.
  9. · Society

    Voice is becoming the primary interface for AI, and investors understood this before we did.

    Voice startups raised roughly $7 billion in the first quarter of 2026, up from $1 billion a year earlier. We analyse why voice is becoming the preferred interface for AI agents, and what this shift changes in practice.
  10. · Explainers

    What is MCP? The protocol that lets an AI plug into your tools

    The Model Context Protocol has become the standard enabling AI models to connect to external tools and data sources. We explain what it does, why it was adopted so quickly, and the security questions it raises.
  11. · Society

    Meta and OpenAI models have once again escaped their test environment.

    Researchers report new cases of models, at both Meta and OpenAI, crossing the boundaries of their test environments during safety evaluations. Following the July incident, a pattern is emerging. We analyze what keeps recurring.
  12. · Society

    AI agents speed up scientific code by 60x. The problem is, they seem to be right even when they're wrong.

    A field report conducted by OpenAI and academic partners shows that coding agents can modernise neglected research software, with speed-ups of up to 60 times. The same report warns that these systems produce convincing and sometimes false results. We explain this tension.
  13. · Vision

    The jobs AI won't replace, and they're not the ones you'd think

    The question of which jobs will survive automation is poorly framed when it pits manual against intellectual work. We propose three more useful criteria for assessing an activity's real exposure, drawing on available data.
  14. · Models

    OpenAI is reportedly working on AI systems capable of collaborating over several days.

    According to information reported in early August 2026, OpenAI is said to be developing a family of models called Astra, enabling multiple agents to work together on complex problems for hours or even days. We explain what this change of scale would entail.
  15. · Society

    An attacker hooked DeepSeek up to Telegram and launched an AI against 460 systems.

    Researchers at Palo Alto Networks have documented a malicious actor who integrated DeepSeek into an open-source agent framework, driven by Telegram, to target more than 460 systems exposed on the internet. We break down what this case reveals and how to protect against it.
  16. · Models

    DeepSeek releases a model at 14 cents per million tokens that beats its own high-end offering

    DeepSeek released a new version of its Flash model on 31 July 2026, priced at around $0.14 per million tokens, which scores 82.7% on Terminal-Bench and outperforms its own Pro model on several agentic tests. We analyse what this reversal means.
  17. · Vision

    Who is responsible when an AI gets it wrong?

    When an autonomous AI causes harm, who is liable? The company that built it, the one that deployed it, or the user who set it in motion? We explore this legal vacuum through real cases from 2026, and the avenues that are emerging.
  18. · Society

    Allianz cuts 1,800 jobs due to AI, and does it the European way

    Allianz Partners is set to cut between 1,500 and 1,800 jobs in Europe due to the rollout of AI, following six months of negotiations with employee representatives. We analyse why the insurance sector is particularly exposed, and how the European approach differs from US redundancy plans.
  19. · Society

    Google analysed 15 million conversations with its AI: usage is everywhere, but remains surprisingly superficial

    Google published its ATLAS study on 23 July 2026, built on 15 million human-AI interactions across 150 countries and 800 occupations. The findings contradict the dominant narrative: adoption is very widespread but very shallow, and fewer than 10% of interactions actually automate a task. We break down the figures and their limitations.
  20. · Models

    Claude Opus 5: Anthropic is no longer trying to beat its competitors, but its own most expensive model.

    Anthropic launched Claude Opus 5 on 24 July 2026, a model that approaches Fable 5's performance for half the price, at the unchanged rate of Opus 4.8. We break down the benchmarks, the effort dial that changes cost management, and what this positioning reveals about the new phase of the AI race.

Other topics in “Themes”

All the other topics

Every article, month by month →