🌏 繁體中文版
After web bugs, memory, virtual machines and TEEs, the headline payouts become easier to read. This map compares representative categories from five programs. The useful question is what outcome a payment requires, and whether you can produce that evidence. The figures below are in US dollars, from official rules available on October 15, 2026.
Workflow position: choose a program → read scope → understand the target → find a bug → report it → weigh the payout and research effort.
What each number actually buys
| Program and representative category | Official amount | Conditions to read alongside it |
|---|---|---|
| Google/Alphabet VRP: server RCE, S0 on T0 or T1 | $101,010 base | Report-quality factors are 0.8, 1 and 1.2, at the panel’s discretion |
| Microsoft Hyper-V: L1 guest RCE | Up to $250,000 | This cell requires a functioning exploit and a high-quality report; reports without an exploit have separate ceilings |
| Meta WhatsApp Private Processing: TEE code execution, privilege escalation, or VM escape | Each up to $300,000 | External attack scenarios; internal scenarios use a different table |
| Apple: network attack requiring no interaction, reaching kernel control | Up to $2,000,000 | The exploit-chain outcome and proof requirements for that category must be met |
| Apple PCC: remote attack on request data achieving specified unsigned code execution | $1,000,000 category | PCC has its own trust boundary and requirements, separate from device-kernel categories |
| Mozilla client security: Highest Impact | Up to $20,000 | Depends on actual impact, supported configurations and report quality; a qualifying sandbox escape is one example |
These categories connect to the preceding lessons. They are not a ranking of every company’s largest possible payment across all products. Some programs also have bonuses and exclusions, so the amounts do not price identical work.
Google: choose the right cell before applying quality
Google’s current rules distinguish target tiers, vulnerability categories and impact. Server RCE in S0 has a $101,010 base for T0 and T1, and $75,000 for normal Google applications in T2. A hostname ending in google.com does not automatically make it T0.
Quality then applies to the appropriate base. If the panel assigns exceptional quality, the T0/S0 example is:
101,010 × 1.2 = $121,212That is an example for a particular cell. A reproducible PoC, impact analysis and clear responses help triage, but checking boxes does not guarantee a 1.2 factor. Other adjustments exist, and the panel decides the final amount. A historical headline maximum also does not establish a universal 1.5 multiplier in the current rules.
Microsoft: exploit evidence and report quality define the cell
The Hyper-V program lists separate ceilings for L1 guest RCE. With a functioning exploit and a high-quality report, the maximum is $250,000. Without an exploit, high-, medium- and low-quality reports have ceilings of $200,000, $100,000 and $50,000 respectively. L2 categories reach up to $150,000; the high-quality information-disclosure category reaches $25,000.
L1 and L2 are virtualization levels. They cannot be reduced to “guest-to-host” versus “guest-to-guest.” Explain the starting privileges, nested environment, boundary crossed and demonstrated outcome. This is why part 10 traces a real case’s data path instead of stopping at “VM escape.”
Google’s 0.8/1/1.2 factors do not apply to this table. Use Microsoft’s eligibility conditions and the ceiling in the relevant cell.
Meta: external and internal describe the threat scenario
The Private Processing guidelines apply to WhatsApp’s named system. In external scenarios, code execution inside the TEE, including bypassing code attestation, privilege escalation and escape to the host each reach up to $300,000. Data leaks reach $150,000 and persistence $10,000.
Internal scenarios use different limits: code execution, privilege escalation and data leaks each reach $50,000; VM escape is N/A and persistence reaches $5,000. A sixfold difference is not a report-quality penalty. The attacker starts in a different position. Nor does this table apply to every Intel SGX, AMD SEV or AI product: first establish that the affected system is in scope.
Apple: distinguish exploit-chain outcomes from PCC
Apple’s categories page organizes rewards around demonstrated outcomes. A network attack requiring no interaction and reaching kernel control has a category up to $2,000,000. Browser-to-kernel control has a $1,000,000 category. Where Target Flags are required, their proof requirements must be met.
PCC has separate conditions. A remote attack on request data that achieves unsigned code execution with arbitrary entitlements within the Apple PCC software image belongs to a $1,000,000 category. Access to user request data outside the trust boundary has a separate $250,000 category. These are not a combined requirement to compromise a device kernel and PCC in one chain.
The program homepage also describes conditional bonuses. Plan around an outcome you can demonstrate, then check each additional condition. Possible bonuses are not guaranteed payments.
Mozilla: make the issue diagnosable and fixable
Mozilla’s client bug bounty has a base table up to $20,000 for valid, potentially exploitable high- or critical-rated client security vulnerabilities, depending on impact and report quality. Highest Impact includes a qualifying sandbox escape. The Higher Impact category containing UXSS reaches $10,000; the High Impact category reaches $3,000.
The report needs enough information to diagnose and fix the issue, a simple reproducible test, and evidence such as an ASan trace or crash dump, or a root-cause analysis. A full exploit is not required. The page separately covers mitigation bypasses and qualifying bonuses: $20,000 is the ceiling in this base table, not an absolute ceiling for every possible program scenario.
Turn the map into a research decision
Larger figures often come with harder demonstration requirements. When comparing opportunities, write down four things: what target you may legally test, the attacker’s starting position, the boundary to cross, and the evidence you can submit. This is a planning method, not a claim about how vendors set prices.
For a beginner, part 4’s local XSS and IDOR exercises already train reproduction and authorization reasoning. A large Hyper-V or kernel-chain award does not mean it should become your first month’s target. Part 6’s time-budget and expected-value framework, followed by the next chapter’s practice plan, connects the price map to action.
Check your understanding
- Can you name the product, category and prerequisites behind each amount?
- Can you distinguish Google quality factors, Microsoft table ceilings and Meta’s external/internal scenarios?
- Why should Apple device exploit chains and PCC be compared separately?
- If you have demonstrated a crash but not a crossed security boundary, how should the report describe it?
- What reproducible evidence will next week’s exercise produce?
Previous: AI product vulnerabilities · Next: certifications and a learning map