Researchers said three attacks against organisations in Japan used AI in different ways, including one campaign that delegated a repeatable workflow to 44 agents inside CyberStrikeAI in October 2026.
Read briefGoogle Deepmind introduced Genie 4, a real-time world model that simulates interactive 3D scenes with sound and dialogue, while Gemini models handle planning and the system remains experimental.
Read briefOracle said it would add ChatGPT and Codex to its applications and widen Oracle Cloud Infrastructure access to Google TPUs, OpenAI's Jalapeño chip and Cerebras hardware.
Read briefGoogle said Genie 4 would generate interactive AI experiences with more realistic worlds, audio and longer simulations, and it would first reach trusted testers as a research preview.
Read briefStephanie Palazzolo / The Information : Anthropic's S-1 shows an 85% gross margin in H1 from running its models, a reality check for skeptics who said pay-as-you-go model sales wouldn't be profitable — When Anthropic on Monday dropped its long-awaited initial public offering…
Read sourceApple said a beta assistant built on Apple and OpenAI models would draft, edit and organise work across Keynote, Pages and Numbers, with third-party agent support on newer Macs.
Read briefThore Graepel's Metis Reasoning proposed an external knowledge-ledger approach to AI reasoning, saying it could reduce hallucinations and produce checkable explanations, but the company described it as research without a working system.
Read briefOpenAI said internal models produced proofs in computer science, mathematics and physics, then asked outside specialists to examine them, with separate review arrangements for each result and some proofs translated into Lean.
Read briefAn analysis said ChatGPT handled partner pages as prepared summaries and ranked them ahead of ordinary web results, while later research found uneven treatment and imperfect citation.
Read briefThe article said generative AI has made phishing pages easier to produce, harder to spot and more adaptable, while also outlining controls meant to limit credential theft and account takeover.
Read briefAnthropic said its IPO filing reiterated that Claude would stay free of advertising, and the company used the letter to frame that choice as a trade-off against revenue and growth.
Read briefOpenAI’s Ted Sanders described how GPT-6.1 Terra fits into the company’s model line-up, saying size matters but does not fully determine performance across different tasks.
Read briefXint said OpenAI's compaction feature could be manipulated to turn a model's hidden reasoning into readable text, and said OpenAI later fixed the issue after a bug bounty report.
Read briefOpenAI and Oracle are expanding their partnership to bring business applications and AI agents into ChatGPT and power new Oracle offerings.
Read sourceCheck Point said it had expanded its work with OpenAI through Daybreak, adding frontier cyber models to its products and workflows and using them to speed vulnerability research and response.
Read briefOpenAI launched GPT-6.1 Terra, a model aimed at computer use, visual reasoning and coding, and said it was cheaper and faster than its earlier Terra and Sol models.
Read briefOpenAI said GPT-6.1 Terra was a mid-range model for routine office work and software control, priced at half the rate of GPT-5.6 Terra and sold through ChatGPT Work, Codex and the API.
Read briefAnthropic filed an S-1 for a Nasdaq listing while reporting $11.6 billion in Q2 2026 revenue, its first GAAP operating profit, and long-term compute commitments of $780 billion to $810 billion.
Read briefOpenAI shares model results on the Kakeya conjecture, Exponential Time Hypothesis, and quantum chromodynamics—and how researchers can verify and build on them.
Read sourceAnthropic filed an S-1 with the Securities and Exchange Commission, disclosing rapid revenue growth, heavy losses and large infrastructure spending as it prepared a proposed public listing.
Read briefExplore OpenAI Signals data on 165 work activities across nearly 100 countries, revealing how people use ChatGPT for work and how usage changes over time.
Read sourceApple expanded its fix for CVE-2026-86950 to iOS 18.7.11 and iPadOS 18.7.11 after saying the CoreGraphics flaw may have been used against targeted individuals on earlier iOS versions.
Read briefMicrosoft patched a flaw in Authenticator for Android that let a malicious app impersonate another Microsoft app and obtain work-account tokens, though the company said exploitation still needed user action.
Read briefKaspersky said the Lazarus Group ran Operation FlashHole against at least three South Korean organisations from July 2025, using targeted phishing, software abuse and post-quantum encryption to load malware in memory.
Read briefApple said it had backported fixes for a CoreGraphics out-of-bounds write flaw to older iPhones and iPads after warning that the bug had been used in targeted attacks.
Read briefThe Australian Signals Directorate said it reached a criminal Amadey backend and severed its link to 180,000 infected machines worldwide, while warning that earlier thefts could still be used.
Read briefENGIE said attackers used a partner-facing marketing tool to reach customer information, limited the incident to France, and notified authorities and affected customers after cutting off access.
Read briefCISA placed two PHP vulnerabilities known as PHP2Shell on its exploited-flaws catalogue after reports of attack activity, while the PHP team said the issues had been patched and earlier proof-of-concepts were not working.
Read briefCISA placed two PHP vulnerabilities in its exploited list after Perplexity said it had seen remote code execution against honeypots, while PHP and the PHP Foundation released fixes.
Read briefThe University of Copenhagen said hackers reached a test environment holding personal data, later added that PIN codes for 20,000 more people were exposed, and reported the incident to Danish regulators.
Read briefSansec said an unknown actor compromised 953 online stores, mainly Magento sites, and used CMS blocks to deliver JavaScript loaders that filtered out scanners and changed domains over time.
Read briefPHP rebuilt its Windows builds for versions 8.2 through 8.5 after a flaw in bundled Brotli code, and said the fixed packages preserved earlier remote code execution patches.
Read briefA buffer overflow in the Brotli compression library, tracked as CVE-2026-111497, affected multiple software stacks, prompted fixes from Google, Node.js and Microsoft, and had no confirmed abuse in the wild.
Read briefArctic Wolf Labs said attackers had used Microsoft Teams external chats, fake IT-support identities and mail bombing to reach victims, gain remote access and in some cases deploy Chaos ransomware.
Read briefMicrosoft released out-of-band security updates for .NET 8, .NET 9, .NET 10 and .NET 11 Release Candidate 3 after disclosing a Brotli decoder flaw that could allow remote code execution.
Read briefHuntress said a token replay campaign ran from July to September 2026, using adversary-in-the-middle login pages and stolen sign-in tokens to compromise Microsoft 365 accounts at 79 organisations.
Read brief