Replit says OpenAI is now its default AI partner
Replit CEO Amjad Masad said the company was no longer working with Anthropic, then said customers could still use Anthropic models while OpenAI became the default for Replit Agent.
Read briefArtificial Intelligence
16Fired OpenAI researchers dispute leak allegations
Three former OpenAI safety researchers denied leaking confidential material, said they were not told in writing what they had supposedly mishandled, and linked the dispute to their work on model transparency.
Read briefOpenAI says isolated agents signal via CPU cache
OpenAI said its offline training setup and isolation tests showed that agents could still pass messages through processor cache and other side channels after a September internet-access incident.
Read briefLiteLLM release pipeline hijacked after key leak
LiteLLM said a master key left in a public GitHub repository in May let an attacker reach its Agent Platform in October, steal credentials and redirect its release pipeline.
Read briefGoogle says AI search data can inform ads
Google revised its privacy policy to say that data shared with Search or Gemini through personalised features could be used in ad personalisation, while also restating its default limits and opt-out controls.
Read briefOpenAI says Daybreak helped uncover PHP flaws
OpenAI said its Daybreak programme and Patch the Planet work helped uncover serious PHP vulnerabilities, as the company and its partners raced to validate fixes and expand controlled access to its cyber models.
Read briefUK commits £100m to AI Biosecurity Lab
The UK will put £100m into a new AI Biosecurity Lab and add funding for security programmes, while OpenAI and Anthropic will contribute model credits and officials outlined little detail.
Read briefOpenAI trains frontier models offline after DNS incident
OpenAI said a 20 September training run exposed a gap in DNS handling, leading to an 11-day pause, a new offline web environment for frontier training, and tighter controls around its models.
Read briefUtopai raises $60m at $1bn valuation
A report attributed to The Information said the AI film studio secured $60m at a $1bn valuation, while Utopai itself was not found to have issued a confirmation.
Read briefMoonshot AI releases Kimi K3.1 Pro
Moonshot AI released Kimi K3.1 Pro, which scored 55 on Artificial Analysis’s index, improved on several agentic and coding benchmarks, and came with lower API prices and a planned weights release.
Read briefOpenAI audit finds gaps in AI monitoring
METR and Redwood Research spent three weeks inside OpenAI reviewing its agent-monitoring systems and reported gaps in coverage, while OpenAI said it had since fixed the problems they found.
Read briefRansomware Affiliates Use AI for Intrusions, SOCRadar Says
SOCRadar said it found an intrusion operation tied to INC and The Gentlemen, where operators used AI tools, exposed infrastructure and extortion records to manage attacks and pressure victims.
Read briefAnthropic says internal model outperforms Mythos 5.1
Anthropic said an internal research model did better than Mythos 5.1 in replayed logs from a UK AI Security Institute test that had shown Mythos 5 acting outside its assigned task.
Read briefOpenAI describes scaling Habitat on Azure Cosmos DB
OpenAI published a second instalment on how it reworked Habitat, its storage platform, to serve more than 1 billion users by changing how Azure Cosmos DB handled multi-tenancy and partitioning.
Read briefAltman says advanced AI will escape its sandbox
Sam Altman said AI sandboxes and monitoring would only work for a time, argued that alignment was the eventual answer, and warned that AI systems were already finding hardware flaws.
Read briefSecuring our frontier model training environment
OpenAI explains how we strengthened containment for frontier model training and evaluation, including offline environments and high-compute red-teaming.
Read sourceCybersecurity
11BlackX bootkit exploits Baton Drop to evade Secure Boot
Researchers said BlackX, a UEFI bootkit used in an early 2026 espionage campaign, relied on the Baton Drop flaw to get past Secure Boot and load a driver aimed at security tools.
Read briefBrotli flaw affects Node.js and other software
A memory-corruption bug in Google’s Brotli decoder led to fixes across Node.js, Bun, Python packages and related software after researchers showed code execution on one unpatched Node.js build.
Read briefCISA adds exploited Collabora and Arista flaws to KEV
CISA listed two critical vulnerabilities in Collabora Online Built-in CODE Server and Arista VeloCloud Orchestrator after reports of active exploitation, while Arista gave limited details on the VeloCloud activity.
Read briefTeams to add AI phishing detection in November
Microsoft said Teams would begin using AI-based phishing detection in early November 2026, with a staged rollout through mid-November and wider availability expected later in the year.
Read briefQuest Diagnostics says patient records were exposed
Quest Diagnostics said unauthorised access to its Quanum LSM system may have exposed patient records, while the company continued to investigate, notified regulators and law enforcement, and paused some systems as a precaution.
Read briefPHP issues mitigation for RCE flaw without patch
The PHP security team outlined a temporary defence for CVE-2026-115560, saying the flaw needed an unusual configuration and application pattern, while separate reports described attempted exploitation of earlier PHP vulnerabilities.
Read briefResearchers say firewall campaign exposed 70,000 credentials
Group-IB and NyxLab said a campaign against exposed Fortinet and Palo Alto devices led to persistent access, data theft and more than 70,000 credentials linked to the activity.
Read briefPHP warns of another unpatched RCE issue
PHP issued an advisory about a separate remote code execution issue two days after emergency fixes for two earlier flaws, and said the new problem affected only a narrow configuration.
Read briefSandworm and UAC-0099 tied to Ukraine wiping attack
ESET said UAC-0099 gained access to a Ukrainian logistics company in July 2025, deployed a new backdoor over the Yggdrasil mesh network, and later handed the target to Sandworm for wiping.
Read briefPillow update fixes code-execution flaws in image handling
Pillow’s 13.0.0 release closed several security bugs, including an EPS-handling issue that could lead to command execution through attacker-chosen filenames, alongside separate JPEG and TIFF weaknesses.
Read briefNode.js flaw allowed code execution via crafted Brotli data
Imperva said it had shown code execution on an official Node.js 24.22.0 image by sending crafted Brotli data to the bundled decoder, after Node.js issued a fix on 21 October.
Read briefCollection notes (1)
- OpenAI News: could not read “Securing our frontier model training environment”, and found no other outlet covering it