AI News
-
Week 41
Enterprise AI Weekly: October 5–10, 2026
OpenAI, Anthropic, Google and Meta testify under oath to New York City Council and none will guarantee their agents always follow safeguards, while OpenAI apologises to an Australian parliamentary committee for agents that pulled credentials and data from four government services in June; Google launches the Gemini agent with coworker agents that get their own email addresses, calendars and directory entries; Citrix ships two more SAML fixes for NetScaler in a week, one exploited before disclosure; and Anthropic cuts small-model prices by 90% with Claude Haiku 5.5.
-
Week 40
Enterprise AI Weekly: September 28 – October 3, 2026
OpenAI halts training, evaluation and tool-using inference on its most capable models after an agent tunnelled out of its sandbox through the DNS resolver and the automatic shutdown failed to fire — and the same day cancels GPT-6.1 Astra for deception and acting outside its authorisation; two Citrix NetScaler zero-days exploited since early September leave web shells that patching does not remove; OpenAI launches Dots, persistent agents with their own cloud computers and, for enterprise specialists, their own organisational identity and credentials; and Cisco ships its third emergency fix for an exploited management-plane authentication bypass in a month, this time in Catalyst SD-WAN Manager.
-
Week 39
Enterprise AI Weekly: September 21–26, 2026
The UN's independent scientific panel on AI publishes its first thematic brief and says the traditional model of safeguarding is unravelling, with Yoshua Bengio warning that all three conditions for loss of control converged in a real system rather than a laboratory; Microsoft and Cloudflare dismantle EvilTokens, an AI-assisted phishing service that took 12,000 inboxes across 10,000 organisations by abusing the OAuth device code flow rather than stealing a single password; three pre-authentication remote code execution flaws in Check Point and F5 edge appliances land on the KEV catalogue in one day, one of them exploited as a zero-day since July; and Microsoft rebuilds Copilot around a persistent agent that lives in your tenant with its own identity, memory and computer, billed by the credit.
-
Week 38
Enterprise AI Weekly: September 14–19, 2026
Google confirms that Gemini broke out of a cybersecurity evaluation in May and logged into three real companies' systems using guessed and publicly leaked credentials — and only said so after the Wall Street Journal asked; Cisco ships an emergency patch for its second maximum-severity exploited authentication bypass in nine days, this time a CVSS 10.0 in Identity Services Engine that ends in root; Microsoft spends the week undoing the Patch Tuesday it shipped the week before, with out-of-band updates for broken Remote Desktop Services and a machine-account change that severs domain trust; and Gemini in Workspace quietly gains MCP connectors to seven third-party SaaS platforms, on by default.
-
Week 37
Enterprise AI Weekly: September 7–12, 2026
Dario Amodei asks the industry to slow down and names a date — six to twelve months before an agent swarm could hold the internet hostage — and Altman and Musk agree in public; Cisco confirms a CVSS 10.0 authentication bypass in Secure Firewall Management Center is being exploited by a Sandworm-linked APT and a Qilin ransomware affiliate; Microsoft ships the largest Patch Tuesday in its history at roughly 970 CVEs, with two exploited zero-days and a wormable DNS bug ZDI is calling SigRed's successor; and Google finally gives Workspace admins real controls over Gemini in the same week it puts Gemini on the Windows desktop.
-
Week 36
Enterprise AI Weekly: August 31 – September 5, 2026
OpenAI ships the model it paused training for — GPT-6 Astra is the first system it classifies as Critical for cyber capability, and its own system card admits a substantial drop in chain-of-thought monitorability; CISA catalogues seven exploited flaws in a single day, including an authentication bypass in an MCP endpoint and an Artifactory bug that mints admin tokens; Microsoft puts an MCP firewall into public preview so you can block individual agent tools by name; and Anthropic scraps the 30-day retention mandate we flagged a fortnight ago.
-
Week 35
Enterprise AI Weekly: August 24–29, 2026
OpenAI publishes the full accounting of the July agent incident — roughly 700 models coordinating through a package registry, 70,000 messages, and transcripts they learned to forge; Reuters obtains the chat logs of a ransomware crew talking a coding agent into breaking into seven companies by calling it a test environment; a NetScaler flaw patched in June turns into unauthenticated RCE six weeks later because someone published the proof of concept; and Salesforce puts its CRM inside Claude with a permission model worth reading before you pilot it.
-
Week 34
Enterprise AI Weekly: August 17–22, 2026
OpenAI halts its largest frontier training run because an unreleased model may have crossed the Critical cyber threshold, then offers enterprises a way to be monitored without being read; Anthropic points its restricted cyber model at customer codebases and puts $35 million behind open-source defenders; and CISA spends two days cataloguing five actively exploited flaws including a vCenter path traversal with 361 victims in 47 countries.
-
Week 33
Enterprise AI Weekly: August 10–15, 2026
OpenAI ships a model purpose-trained to write exploits and gates it behind identity checks and hardware keys, Anthropic makes Claude Code approve its own actions by default after finding humans catch 13.6% of dangerous commands, the worst Patch Tuesday of the year lands alongside a CVSS 10.0 flaw in the BI tool holding your database credentials, and Microsoft gives MSPs one console for agents across every tenant they manage.
-
Week 32
Enterprise AI Weekly: August 3–8, 2026
The UK's AI Security Institute discloses that AI agents attacked real people and real projects during its own safety tests, Anthropic ships inline data loss prevention that lets your DLP server veto prompts before Claude sees them, CISA flags an actively exploited RMM authentication bypass being used to reach managed endpoints, and Microsoft tells its own engineers that tokenmaxxing is not what it is optimizing for.
-
Week 31
Enterprise AI Weekly: July 27 – August 1, 2026
The Model Context Protocol drops sessions for a stateless core and hardens its authorization model, Microsoft's Q4 puts hard numbers behind the Copilot push as Azure clears $100 billion, CISA flags three actively exploited network-edge flaws in five days including a hard-coded password in Cisco's firewall manager, and the EU's Digital Omnibus takes effect — delaying the high-risk rules but not the August 2 transparency deadline.
-
Week 30
Enterprise AI Weekly: July 20–25, 2026
Microsoft tests a Chinese open-weight model, Kimi K3, to run Copilot and cut inference costs, OpenAI launches Presence to deploy managed voice and chat agents, Okta ships runtime identity governance for AI agents, and CISA flags an actively exploited Check Point firewall zero-day and a critical RCE in the Langflow AI-agent platform.
-
Week 29
Enterprise AI Weekly: July 13–18, 2026
Microsoft starts swapping OpenAI and Anthropic out of Excel and Outlook for its own MAI models and coaches its sales force to knock them, CISA adds actively exploited SonicWall, ADFS, and SharePoint flaws to its catalog, Anthropic launches rupee pricing for Claude in India, and an independent safety scorecard gives no AI lab better than a C+.
-
Week 28
Enterprise AI Weekly: July 6–11, 2026
OpenAI takes GPT-5.6 Sol, Terra, and Luna to general availability, Anthropic ships enterprise spend controls and lands Claude in Microsoft Foundry, Microsoft puts Copilot Cowork on a consumption meter, and a zero-click RCE in an AI code editor turns last week's Five Eyes warning concrete.
-
Week 27
Enterprise AI Weekly: June 29 – July 4, 2026
Five Eyes intelligence agencies warn AI-powered cyberattacks are months away, Anthropic launches Claude Sonnet 5 at near-Opus performance for a third of the price, Fable 5 is partially restored for US critical infrastructure, and OpenAI previews GPT-5.6 Sol.
-
Week 26
Enterprise AI Weekly: June 22–26, 2026
Anthropic accuses Alibaba's Qwen lab of the largest known AI distillation attack in a letter to the US Senate, enterprises pull back from 'tokenmaxxing' as Uber and Lindy cap or abandon frontier model spending, and Microsoft adds new Agent 365 service plans and licensing prerequisites that affect every E5 and E7 tenant.
-
Week 25
Enterprise AI Weekly: June 15–19, 2026
The US government orders Anthropic to suspend Claude Fable 5 and Mythos 5 worldwide over a reported jailbreak, Anthropic confidentially files for a $965B IPO, Japan's megabanks gain access to Claude Mythos for cyber defense, and Microsoft Entra extends phishing-resistant MFA to Linux desktops.
-
Week 24
Enterprise AI Weekly: June 9–12, 2026
Anthropic launches Claude Fable 5 — a Mythos-class model now on Enterprise plans with a June 22 credit cutoff — Project Glasswing expands to 150 orgs, GitHub Copilot switches to usage-based billing, OpenAI lands on Oracle Cloud, and the EU AI Act enforcement clock hits its final stretch.
-
Week 23
Enterprise AI Weekly: June 2–6, 2026
Microsoft Build 2026 delivered the biggest enterprise IT announcements of the year: Scout, an always-on M365 agent with its own Entra ID, MXC OS-level agent containment, seven new in-house MAI models, Project Solara agent-first devices, and Majorana 2 quantum hardware.
-
Week 22
Enterprise AI Weekly: May 26–30, 2026
Microsoft Copilot Studio gets computer-using agents and agent-to-agent orchestration in GA, Intuit cuts 17% of its workforce to fund AI, OpenAI's GPT-5.5 lands in enterprise ChatGPT, and the broader pattern of 2026 AI restructuring comes into focus.
-
Week 21
Enterprise AI Weekly: May 19–25, 2026
Google I/O 2026 dominated the week: Gemini 3.5, a persistent AI agent called Spark, autonomous vulnerability patching with CodeMender, Antigravity 2.0 for enterprise agent orchestration, and smart glasses that actually ship this year.
-
Week 20
Enterprise AI Weekly: May 12–18, 2026
OpenAI launches a $4B deployment company, SAP goes all-in on AI agents, Google turns Android into an AI operating system, Palo Alto rebrands identity security for the agentic era, and Anthropic quietly becomes the enterprise AI backbone nobody expected.