Microsoft patches 1,259 flaws including exploited Outlook zero day as Railway discloses 740,000 dollar AI key breach

Enterprise security teams confronted a twin wave of urgent patching and cloud-credential remediation on Wednesday after Microsoft released fixes for a massive 1,259 software vulnerabilities — headlined by an actively exploited Microsoft Outlook zero-day that silently leaks Windows network credentials — while developer cloud platform Railway disclosed that intruders stole stored customer artificial-intelligence API keys to run up an estimated $740,000 in unauthorized compute charges.

Microsoft’s November Patch Tuesday bundle marks the second consecutive month in which Redmond’s monthly vulnerability tally has topped 1,000 flaws, pushing Microsoft past 5,100 remediated CVEs in 2026 alone as AI-assisted vulnerability discovery accelerates bug reporting across legacy and cloud codebases.

At the same time, Railway’s incident report offered one of the clearest financial accounting disclosures to date of large-scale “LLMjacking,” arriving just days after an infrastructure breach at inference provider Fireworks AI underscored escalating attacks on cloud AI secrets.

Actively exploited Outlook zero-day CVE-2026-124194 triggers CISA mandate

Of the 1,259 flaws addressed across Microsoft’s portfolio — including 726 vulnerabilities in Windows, 174 in Microsoft Office, 59 in SQL Server, and 27 in SharePoint Server, with 63 rated Critical — defenders’ highest priority is CVE-2026-124194, an Outlook spoofing vulnerability carrying a CVSS score of 6.5 that Microsoft confirmed is already under active exploitation in the wild.

According to Microsoft’s security advisory and vulnerability researchers at Tenable, an unauthenticated attacker can trigger the flaw by sending a specially crafted message file to a target running affected editions of Microsoft 365 Apps for Enterprise or Office LTSC 2021 and 2024. When opened in Outlook, a user-interface flaw bypasses standard external-content safety prompts, causing the client to initiate an outbound connection to an attacker-controlled server and expose the victim’s NTLMv2 password hash for offline cracking or relay attacks.

The US Cybersecurity and Infrastructure Security Agency added CVE-2026-124194 to its Known Exploited Vulnerabilities catalog, ordering federal civilian executive-branch agencies to deploy Microsoft’s update no later than November 24 and urging private-sector enterprises to block outbound SMB traffic at network perimeters immediately.

Beyond Outlook, November’s update resolves more than 500 elevation-of-privilege bugs alongside dozens of critical remote-code-execution flaws spanning Windows kernel components, Hyper-V, and developer toolchains.

Attackers decrypted AI provider secrets on Railway for $740,000 compute spree

In a separate post-mortem that rattled software engineering teams, platform-as-a-service provider Railway revealed that attackers exploited infrastructure vulnerabilities during an October intrusion to query and decrypt customer environment variables — specifically hunting for commercial AI model API keys.

Railway estimated that at most 1.85 percent of platform workspaces — an upper bound of roughly 100,000 workspaces — could have been exposed during the intrusion, alongside linked bring-your-own-key agent tokens and GitHub OAuth credentials belonging to 1,852 users. Railway said it immediately invalidated user sessions, rotated affected OAuth tokens, temporarily halted container builds, and found no evidence that attackers accessed customer source-code repositories or production databases.

Instead, the intruders monetized stolen Anthropic, OpenAI, and cloud-inference API keys directly, burning through an estimated $740,000 in unauthorized token generation consistent with black-market API resale or automated model-distillation scraping.

Reversing an initial October support notice that alarmed affected startups, Railway announced it is working directly with major AI model providers to waive and refund fraudulent usage bills — and pledged to cover out of pocket any verified unauthorized AI charges that upstream providers decline to refund. Security architects warned Wednesday that developers should treat frontier AI API keys with the same hardware-backed secret rotation, strict per-day spend caps, and IP allowlisting applied to production payment gateways.