Microsoft Cuts Jobs, Builds Cheaper Mythos Rival

Microsoft doesn't have a model that finds zero-days the way Claude Mythos does. What it has instead, according to reporting from The Information, is a plan to route the same job across three vendors' models and charge less for the result.

The product is codenamed Project Perception. It hasn't shipped yet, and Microsoft hasn't confirmed it publicly. But the restructuring behind it is already visible in the org chart, and it tells you more about Microsoft's strategy than a launch date would.

The Executive Reset Came First

Charlie Bell ran Microsoft Security for three and a half years before moving to an individual-contributor role in February 2026, a shift Microsoft framed as letting him focus on "engineering quality" rather than management. His replacement, Hayete Gallot, is a Microsoft veteran who spent 15 years at the company before leaving for Google Cloud, where she was president of customer experience.

Gallot reports directly to CEO Satya Nadella. Since taking the job, she has replaced at least eight of the executives who used to report to Bell and cut several hundred roles inside a security organization that numbered roughly 10,000 people, according to reporting from The Information picked up by PYMNTS and Winbuzzer. Microsoft has declined to confirm the exact headcount reduction.

"The entire industry is getting reimagined from the ground up. And it will reward the companies that see the shift early, make the hard choices, and execute with discipline."

That's Gallot's own framing of the cuts, and it lines up with what she's building toward: fewer people running manual security workflows, more spend routed into AI-assisted ones.

Microsoft's Redmond campus. The security division Gallot now runs employed roughly 10,000 people before this year's cuts. Source: wikimedia.org

Eight Executives, One Reporting Line

The consolidation isn't just about cost. Gallot has folded engineering teams that used to sit under separate leaders into a smaller reporting structure directly under her, which is the kind of move that makes sense if you're about to ship one flagship product rather than maintain several independent ones. Project Perception looks like that product.

What Project Perception Actually Does

Based on the reporting so far, Project Perception is not a new model. It's an orchestration layer that sits on top of models Microsoft already has access to: its own, plus OpenAI's and Anthropic's through existing commercial deals. The pitch is that a security scan doesn't need a frontier model for every step, so don't pay for one.

The Routing Logic

Routine work such as inventory checks, log parsing, and CVE triage gets handed to cheaper models. Harder problems, like reconstructing an exploit chain or interpreting an authentication flow, get escalated to frontier-tier models from Anthropic or OpenAI. The rough shape of that decision, as described by Windows News, looks something like this:

if task in [inventory_scan, log_triage, cve_match]:
    route(model="microsoft-internal", tier="low_cost")
elif task in [exploit_chain_analysis, auth_flow_reasoning]:
    route(model="frontier", provider=best_fit(anthropic, openai))
else:
    route(model="microsoft-internal", tier="mid")

That's a simplification of what's been described, not a leaked snippet, but it captures the design goal: keep the expensive models on the payroll only for the fraction of work that actually needs them.

Scope: Code, Cloud, and Endpoints

Project Perception is meant to scan a company's source code, cloud configuration, and endpoints together, flag exploitable weaknesses, explain the impact in plain language, and propose a fix. That's the same territory Anthropic staked out with Project Glasswing and OpenAI covers with Codex Security, the product formerly called Aardvark. What's different is that Microsoft isn't trying to out-perform either model on raw capability. It's trying to make continuous scanning cheap enough that customers run it constantly instead of on a schedule.

How the Field Lines Up on Price

Anthropic and OpenAI have both kept their frontier security models behind restricted access and premium pricing, which is exactly the gap Microsoft is aiming at.

Mythos 5 already cut its own predecessor's price in half, from $25/$125 per million tokens down to $10/$50, and it's still the most expensive option on this list by a wide margin. Undercutting it isn't a high bar. Undercutting it while matching what Mythos found, including a 27-year-old OpenBSD bug and a working remote code execution chain against a fully patched FreeBSD NFS server, is a different question entirely.

Where It Falls Short

Nothing about Project Perception has been benchmarked, because nothing about Project Perception has shipped. Every detail so far comes from reporting on internal plans, not from a product page, a technical report, or an eval score. Microsoft has confirmed neither the launch date nor the pricing that multiple outlets have described as "far lower than Mythos."

There's also a structural question the routing pitch doesn't answer. Mythos 5's headline result on ExploitBench, an autonomous exploit-development benchmark, comes from using one very capable model for the whole task, not from splitting the work between a cheap triage pass and an expensive escalation. If the hardest 10% of vulnerability-hunting work still requires a frontier model at frontier prices, Project Perception's savings only apply to the easy 90% that a competent scanner already catches.

The layoffs raise a separate concern. Cutting several hundred roles while consolidating engineering teams under a new reporting line is a fast way to lose the institutional knowledge that makes a security org good at triage in the first place, right as that team is supposed to be shipping a flagship product.

Microsoft hasn't set a launch date for Project Perception. The Information's July 17 report puts a release "as early as this month," which gives the company under two weeks to prove any of this works.

Sources:

• Microsoft's New Security Chief Replaces Top Execs to Force an AI Overhaul - PYMNTS
• Microsoft's AI Cybersecurity Overhaul Brings Hundreds of Layoffs - Winbuzzer
• Microsoft's Project Perception Aims to Make AI Vulnerability Fixing Cheap Enough to Run Nonstop - Windows News
• Microsoft to launch Project Perception AI cybersecurity platform this month - NewsBytes
• OpenAI Unveils Aardvark: GPT-5 Agent That Finds and Fixes Code Flaws Automatically - The Hacker News