Timeline lokal

AI is exposing security holes across the web, and Cloudflare and Vercel executives explain the risks. They say the internet could actually become safer.
Source: Business Insider
https://www.businessinsider.com/cloudflare-vercel-leaders-caution-against-ai-cyber-panic-2026-10
#AI

Companies risk a CFO talent crisis by automating junior finance roles. The work AI does best is the work nobody wants to do. In finance, that is precisely the problem. Automating entry-level tasks removes a crucial stage of training for future finance chiefs.
Source: Fortune
https://fortune.com/2026/10/23/companies-risk-a-cfo-talent-crisis-by-automating-junior-finance-roles/
#EnterpriseAI

The Bank of Japan warned of risks from an AI market shock and cyberattacks, while noting that rising property and stock prices warrant careful attention given still-low interest rates.
Source: The Business Times Singapore
https://www.businesstimes.com.sg/international/bank-japan-warns-risks-ai-market-shock-cyberattacks
#TechPolicy

Researchers introduced an AI surrogate modelling framework that replicates national wildfire hazard maps in seconds, achieving an 806-fold CPU speed increase while using only 13 % of the original simulation memory.
Source: arXiv
https://arxiv.org/abs/2610.30170
#MachineLearning

Robinhood Ventures has invested $10 million in AI coding tool maker Cognition. The investment was part of Cognition's Series E funding round in September. Robinhood Ventures previously backed OpenAI with a $75 million investment in April.
Source: Fortune
https://fortune.com/2026/10/23/exclusive-robinhood-bets-10-million-ai-coding-tool-maker-cognition/
#Startups #OpenAI

When an LLM agent's move fails, writing the move and the failure message into the transcript makes small models more likely to repeat the failed action in tool calling.
Source: arXiv
https://arxiv.org/abs/2610.30205
#MachineLearning

Li et al. introduce EdenGNN-Uni, a universal charge density model for the periodic table that is over an order of magnitude faster than some existing methods, facilitating efficient and accurate electronic structure calculations of materials.
Source: Nature Machine Intelligence
https://www.nature.com/articles/s42256-026-01326-5
#MachineLearning

The paper proposes a Three-Pillar governance framework to make AI agents safe and trustworthy. It emphasizes transparency, accountability and trustworthiness as core pillars.
Source: arXiv
https://arxiv.org/abs/2610.30445
#MachineLearning

Researchers asked whether safety probe performance can be predicted in advance from model size and the number of labels.
Source: arXiv
https://arxiv.org/abs/2610.30678
#MachineLearning

Large language models exhibit systematic failures when incorporating evolving medical knowledge into clinical recommendations according to new research.
Source: arXiv
https://arxiv.org/abs/2610.30631
#MachineLearning

Researchers introduced MemPABench to evaluate memory in personal agents. Current memory systems show limited gains in preference selection and struggle to adapt to preference changes.
Source: arXiv
https://arxiv.org/abs/2610.30485
#MachineLearning

The paper introduces Route-and-Generate, a hierarchical training framework that lets large language models decide how much reasoning to apply for each input.
Source: arXiv
https://arxiv.org/abs/2610.30351
#MachineLearning

ARATest is a property-based testing library that checks property skeletons as executable tests.
Source: arXiv
https://arxiv.org/abs/2610.30495
#MachineLearning

RoleFlow introduces self-evolving multi-role agentic collaboration through structural adaptation. It refines collaboration structures via feedback-driven atomic edits.
Source: arXiv
https://arxiv.org/abs/2610.30429
#MachineLearning

AI control protocols aim to get useful work from models that may be misaligned by monitoring their behaviour.
Source: arXiv
https://arxiv.org/abs/2610.30582
#MachineLearning

An LLM agent can retrieve a memory correctly yet use it incorrectly after its validity state changes. We study governance-state corruption faults that alter which retained records enter model-visible context.
Source: arXiv
https://arxiv.org/abs/2610.30698
#MachineLearning

LiteLLM reported that hackers poisoned its software using GitHub tokens leaked via Claude Code. According to the company, attackers took keys from its own logs to publish a poisoned version on Wednesday.
Source: The Next Web
https://thenextweb.com/news/litellm-hack-github-tokens-logs-claude-code
#AI #Claude

Walmart will let AI agents shop for customers on Walmart.com and at Sam's Club, each with its own profile, spending limit and purchase history.
Source: The Next Web
https://thenextweb.com/news/walmart-ai-agents-shop-agent-profiles
#AIAgents

Anthropic launched OSS Scanner, a free, opt-in service to find vulnerabilities in open-source projects.
Source: InfoQ
https://www.infoq.com/news/2026/10/anthropic-oss-security-scanner/?utm_campaign=infoq_content&utm_source=infoq&utm_medium=feed&utm_term=global
#AI #Anthropic

Fitch Ratings states that AI-related business disruption poses moderate risk to financial institutions over the coming five years.
Source: EconomyNext
https://economynext.com/financial-institutions-face-moderate-business-disruption-risk-from-ai-fitch-287664/
#EnterpriseAI