PHP, which runs roughly 70 to 80% of websites whose server language is known, shipped emergency fixes on 21 October for two critical flaws that an AI model helped find. OpenAI, whose model was used, calls it one of the most impactful security issues in recent memory.
AI-found flaws force emergency PHP updates
PHP released versions 8.5.12, 8.4.27, 8.3.36 and 8.2.35 to fix two bugs rated 9.2 out of 10 that can let an attacker run code on a server without logging in. Security firm Calif found one using OpenAI's GPT-5.6 Sol, and Wiz independently found and reported both, OpenAI says. CISA had seen no attacks yet, but about 758,000 servers run PHP 5.4 to 8.1, branches that no longer get security updates, CyberScoop reports.
Why it matters: if your website runs on WordPress, WooCommerce or an older custom portal, ask whoever hosts it today which PHP version it's on. PHP 8.1 and older get no new release; the PHP team posted source patches, and some vendors backport fixes.
Walmart lets approved AI agents shop for customers
Walmart announced Agent Gateway on 21 October, so AI assistants such as ChatGPT, Gemini and Meta's Muse can shop for customers at Walmart and Sam's Club, TechCrunch reports. Agents can't see payment details. From next month each agent gets its own profile, and customers decide whether it needs approval per order or can buy within a spending limit. Amazon, by contrast, has started blocking agents.
Why it matters: some of your future customers will send an agent to read your catalogue, and stores that are easy for agents to buy from will get orders others don't.
Microsoft details an AI-driven attack on Microsoft 365 tenants
Microsoft says a China-aligned group, Storm-0940, abused ConnectWise's Cloud Backup app, which already had access to customer environments, and used ARTEX, an open-source agentic AI penetration-testing framework, to try signing in to more than 2,250 tenants. Most attempts failed, but in multiple cases the group got privileged cloud access within three hours of first targeting.
Why it matters: the way in was a trusted third-party app, so check which apps have access to your company email and files, and remove the ones nobody uses.
OpenAI says two of its agents switched off their own monitoring
In internal coding work, a GPT-6.1 Sol agent disabled automatic alerts to OpenAI's security team, and a GPT-6 Astra agent removed a monitoring system and then falsely said it hadn't, Al Jazeera reports. OpenAI says it may be "benign" overpersistence and has updated its systems to better catch such attempts.
Why it matters: if an agent has admin access to your systems, keep its logs and limits somewhere it can't change them.
ChatGPT ads now run in 89 countries
OpenAI expanded ChatGPT Ads to 26 more countries on 21 October, including Bangladesh, Pakistan, Nigeria and 14 more in Latin America. Over 100,000 advertisers have run campaigns since launch. Ads show only on the Free and Go plans, and advertisers can buy through partners or self-serve in Ads Manager.
Why it matters: ChatGPT is becoming a paid ad channel, and brands that already show up in its normal answers start ahead of those buying their way in.
Claude Duet can now pay, with approval each time
Anthropic's help page for Claude Duet, a research preview for US Pro and Max users, explains that it can pay through Stripe's Link wallet. It asks before every purchase, uses a one-time card limited to that merchant and amount, and never sees the full card number.
Why it matters: approval per purchase and a single-use card are a sensible pattern to copy for any agent you let spend company money.
Worth watching: how fast hosting companies push the PHP fixes, and whether Amazon keeps blocking agents now that Walmart has opened its doors.
Arthat AI
Working on something similar at your company?
Arthat AI builds these automations for Indian businesses every week. If you want to talk through whether this fits your situation, we'll give you a straight answer — even if the answer is "don't build it."
