A bug bounty program pays researchers for valid, impactful security findings — a distinction CISA's own Binding Operational Directive 20-01 requires every federal agency to draw explicitly; a vulnerability disclosure policy (VDP) is the free, always-on, safe-harbor-protected channel every device maker should already have regardless of whether it ever pays a cent. CISA's own federal directive makes that distinction explicit rather than leaving it to convention, and it matters most for hardware and IoT vendors specifically, because scoping a physical device — firmware layers, companion apps, cloud backends, out-of-band management hardware — is harder to get right than scoping a web application, and getting it wrong has produced real, publicly documented disputes.
Key takeaways
- The VDP comes first, and the bounty is an optional layer on top of it.
- All three require a published way to report vulnerabilities; none requires paying anyone.
- A hardware or IoT program has to decide, layer by layer, what is fair game: device models, firmware versions, physical attacks, companion apps and cloud backends.
- Either they buy it, you register the units they own, or you lend them devices; published programs use all three, and your terms should say which applies.
What comes first, a VDP or a bug bounty?
The VDP comes first, and the bounty is an optional layer on top of it. CISA's Binding Operational Directive 20-01 (2 September 2020), which requires every US federal civilian executive-branch agency to publish a vulnerability disclosure policy, states the distinction directly: "a VDP is similar to, but distinct from, a 'bug bounty.' In bug bounty programs, organizations pay for valid and impactful findings," and the directive "does not require agencies to establish bug bounty programs."
The EU Cyber Resilience Act draws the same line for device makers. Its Annex I Part II requires manufacturers to "put in place and enforce a policy on coordinated vulnerability disclosure" (point 5) and to provide "a contact address for the reporting of the vulnerabilities discovered" (point 6). Paying is optional: recital (76) says manufacturers "should be able to use programmes, as part of their coordinated vulnerability disclosure policies, to incentivise the reporting of vulnerabilities by ensuring that individuals or entities receive recognition and compensation for their efforts. This refers to so-called 'bug bounty programmes'."
A VDP with no reward still gives you a monitored contact, triage and safe harbor, which the UK PSTI regime also expects. A bounty without that process has nothing to plug into — see how to set up a PSIRT for a hardware product company and the vulnerability disclosure policy template for building that baseline first.
What do the CRA, ETSI EN 303 645 and UK PSTI require, and where does a bounty fit?
All three require a published way to report vulnerabilities; none requires paying anyone.
| Source | What it requires | VDP covers it? | What a bounty adds |
|---|---|---|---|
| CRA Annex I Part II (5), (6) | Enforced CVD policy; contact address for reporting | Yes | Optional rewards, per recital (76) |
| CRA Article 13(17), Annex II point 2 | A single point of contact that "shall not limit such means to automated tools" | Yes | A platform form alone does not satisfy it |
| CRA Annex I Part II (4) | Publicly disclose fixed vulnerabilities | Yes, with advisories | Payout terms must not block it |
| ETSI EN 303 645 V3.1.3 5.2-1 | Public VDP with contact and acknowledgement/update timelines | Yes | Nothing required |
| UK SI 2023/1007 Schedule 1 para 2 | Point of contact and acknowledgement/update timing, free and without requesting personal information | Yes | A sign-up-first bounty cannot be the only route |
| ISO/IEC 29147:2018 clause 9 | Policy elements incl. scope (9.3.5), recognition (9.3.7), legal considerations (9.4.2), per the published contents | Yes | Recognition may be monetary |
The timeline matters for hardware. ETSI's supporting text for provision 5.2-2 expects a software fix "within 90 days" and notes that "a hardware fix can take considerably longer", which is worth stating in bounty terms so researchers know a disclosure date may move for a board respin. The CRA dates in Article 71(2) also apply: Article 14 reporting has applied since 11 September 2026 and the rest of the regulation applies from 11 December 2027.
What does scoping a hardware or IoT program actually require?
A hardware or IoT program has to decide, layer by layer, what is fair game: device models, firmware versions, physical attacks, companion apps and cloud backends. See how real published programs, such as VeSync's IoT Bug Bounty Program, draw the line:
| Program | What's explicitly in scope | What's explicitly out |
|---|---|---|
| Intel Vulnerability Disclosure Program (via Intigriti; every asset type listed as "No bounty") | OS-level kernel exploits; out-of-band management hardware (BMC, CSME) | Firmware weaknesses requiring physical access or a UEFI shell; third-party products; pre-release products |
| VeSync IoT Bug Bounty | The VeSync app plus "IoT devices supported by Vesync Application, including outlets, switches, lighting, kitchen appliances, air purifiers, scales, trackers" | Device categories VeSync doesn't make or support |
| Motorola Mobility (via Bugcrowd) | Motorola devices on specific Android versions and current security patch levels | Web-app-related findings |
| Arm Bug Bounty (via Intigriti) | Mali Command Stream Frontend firmware and the Mali GPU kernel driver (Kbase) | "Attacks requiring physical access to a victim's computer/device"; the Linux kernel itself; other Mali drivers |
Scope names models, firmware layers, physical-access methods and companion software, not "all our hardware". Intel's listing shows the same rules can run without any reward. Three calls to make in writing:
Physical attacksIs opening the case, a UART console or an unlocked JTAG/SWD debug port in scope? If your threat model includes stolen or resold devices, say which physical findings you want. Companion app and cloudIs the app in scope? Is the cloud API, which may be run by a third party? VersionsCurrent firmware only, or older releases still supported? End-of-life products are commonly out, as on Intel's listing.
How do researchers get the hardware to test?
Either they buy it, you register the units they own, or you lend them devices; published programs use all three, and your terms should say which applies.
| Route | Published example | Trade-off for a device maker |
|---|---|---|
| Researcher buys the device | Intel: "All testing resources must be obtained by participating security researchers"; Arm: "unable to assist with acquiring devices for testing" | No logistics, but fewer researchers look at expensive or region-locked products |
| Register researcher-owned hardware | Tesla on Bugcrowd: owners researching vehicles or Powerwall "must first register the hardware with Tesla", which "permits Tesla to voluntarily support researchers by providing updates or a 'reflash'" | Lets you restore bricked units and avoid warranty disputes; needs a registration contact |
| Loan dedicated research units | Apple Security Research Device: "a specially fused iPhone" provided "as a 12-month renewable loan", and findings on it are "automatically considered for Apple Security Bounty" | Deepest access for researchers; you build, track and recover fused units |
A small maker can send retail units to a few vetted researchers and promise to reflash or replace bricked ones.
What does safe harbor cover, and what can it never cover?
Safe harbor is a promise not to sue or report a researcher who follows the rules, and it only binds the organisation that makes it. Every published program this guide reviewed limits that protection to systems the operator controls. Arm's program (via Intigriti) says it "will not pursue civil action or initiate a complaint for accidental, good faith violations", but "is not able to grant authority or safe harbor to researchers for any security vulnerability testing of third-party software or systems." Intel's waiver "does not apply to your security research that involves the networks, systems, information, applications, devices, products, or services of another party (which is not Intel)."
The widely reused disclose.io terms (CC0) spell out the same limit: the safe harbor "applies only to legal claims under the control of the organization participating in this policy", and "the policy does not bind independent third parties."
Two US texts show what public law adds, and what it doesn't:
- Computer misuse. The US Justice Manual section on the CFAA (9-48.000) tells prosecutors to "decline prosecution if available evidence shows the defendant's conduct consisted of, and the defendant intended, good-faith security research", defined around avoiding harm and using the information "to promote the security or safety of the class of devices, machines, or online services".
- Anti-circumvention. The DMCA exemption in 37 CFR 201.40(b)(18) covers circumvention "on a lawfully acquired device or machine" for good-faith security research, but states that eligibility "is not a safe harbor from, or defense to, liability under other applicable laws".
Neither replaces your written authorisation, and both are US-only.
How much do IoT and hardware programs pay?
Published consumer-IoT tables top out in the low thousands of dollars per finding; platform vendors pay far more. The figures below are copied from the programs' own pages.
| Program | Published range or maximum (USD) |
|---|---|
| VeSync IoT Bug Bounty | Critical $3,000–$10,000; High $1,000–$3,000; Medium $500–$1,000; Low $100–$500 |
| Arm | Tier 2 $1,000–$20,000; Tier 3 $500–$10,000 |
| Apple Security Bounty | Up to $500,000 for "unauthorized physical device access" to sensitive user data; up to $2,000,000 for a network attack with no user interaction reaching the kernel |
| Intel VDP via Intigriti | No bounty; reputation points only |
VeSync also publishes response targets next to its rewards — "Acknowledged within 2 business days", updates "every 2–3 weeks" — covering the ETSI and PSTI timeline duties in one place.
What happens when scope and terms are unclear? DJI, 2017 and 2026
DJI's 2017 program shows how a bounty fails when rules arrive after the findings, and its 2026 Romo case shows the same company handling a report without a public dispute. Per SecurityWeek's 2017 report, DJI offered "between $100 and $30,000" but "had failed to clarify exactly which of its assets were in scope." Researcher Kevin Finisterre found SSL and firmware AES keys exposed in public source code and "was informed that he qualified for the maximum reward, $30,000", but to receive it "he would have to sign an agreement." After lawyers told him the agreement was "extremely risky" and "likely crafted in bad faith to silence anyone that signed it," he walked away and published his findings.
Nine years later, DroneDJ reported on 10 March 2026 that DJI had promised a $30,000 reward to software engineer Sammy Azdoufal for a cloud authorisation flaw in its Romo robot vacuum backend that "could have allowed access to a fleet of roughly 7,000 devices across 24 countries", and that "two independent security researchers later reported the same vulnerability through the company's bug bounty program." The reward ceiling was the same in both years; what differed was whether scope, release terms and the payment process existed before the report. See ISO/IEC 29147 and 30111 and the FIRST PSIRT Services Framework for the intake process underneath.
Worked example: what would a smart-plug maker's first bounty brief look like?
Synthetic example. The company, products and amounts are invented for illustration and are not a SoftWSP customer.
A maker of a Wi-Fi smart plug and its app already runs a VDP with a published security.txt (generated with the security.txt generator) and a two-person PSIRT. It decides to add a private bounty for 20 invited researchers for six months.
ScopePlug models P1 and P2 on the two latest firmware releases, the app and the device cloud API. Out: the payment provider it does not operate. Physical findingsIn scope, but only for key extraction or signature bypass that affects other units; a debug port that only exposes the tester's own unit is rated Low. HardwareEach researcher receives two retail units; bricked units are replaced on request. RewardsA fixed severity table near published IoT ranges (say $3,000 for remote takeover, $250 for Low); duplicates pay only the first report. Terms published before launchSafe harbor in disclose.io wording with the third-party limit; disclosure after the fix or 90 days for software, longer for hardware respins; nothing blocks publication after a fix. Exit checkAfter six months, compare valid reports and fix times with the VDP-only period before deciding on a public program.
How does a paid report flow into the PSIRT and CRA reporting?
A bounty report goes through exactly the same steps as any VDP report; only the payout step is extra.
- Acknowledge within the published time and confirm the report is in scope.
- Triage and score severity with CVSS; check whether the component is already known-vulnerable in your SBOM.
- If the flaw is being actively exploited, start the CRA Article 14 early warning (24 hours), notification (72 hours) and later report; see CRA vulnerability reporting in 24 hours.
- Fix, assign a CVE, publish the advisory as Annex I Part II (4) requires, then pay and credit the researcher.
Target times for each step are compared in PSIRT response SLA examples.
What are the common pitfalls?
Hardware bounties usually break on terms missing at launch, not on the amount paid.
- Bounty before VDP. It inherits every gap in the disclosure process, plus payment disputes.
- Platform form as the only channel. The CRA single point of contact may not rely only on automated tools, and PSTI forbids asking for personal information as a condition of reporting.
- Safe harbor that overpromises. Name the third-party firmware and backends it does not cover.
- Release terms written after the finding, as in DJI's 2017 dispute.
- No hardware plan. Expensive products get little attention and bricked units become disputes.
Does AI triage bug bounty submissions?
No. SoftWSP does not read, score, deduplicate or price researcher submissions; severity, validity, scope and payout stay with your PSIRT. What it does is narrower and upstream: from a firmware image it extracts an SBOM (CycloneDX 1.6), matches the components against NVD, CISA KEV, FIRST EPSS and CISA CSAF ICS advisories, and drafts a cited report that an engineer reviews. When a bounty report names a third-party component, that report tells the reviewer whether the image you scanned ships it and which public advisories already cover it, and a watchlist alert on that CVE or vendor/product flags later records. It does not show whether the vulnerable code is reachable, and it does not write VEX statements.
Paying for findings can surface issues a free VDP alone would miss, but only once the disclosure process, scope and safe-harbor terms are solid — the foundation a PSIRT program and a published VDP provide. SoftWSP's PSIRT service builds that intake, triage and advisory process first, so a paid bounty layer — if you choose to add one — has something solid to sit on: see PSIRT.
This guide draws on CISA's federal directive, the EU Cyber Resilience Act, ETSI EN 303 645, the UK PSTI regulations, the ISO/IEC 29147 contents, US federal texts on good-faith research, published program pages (Intigriti, Bugcrowd, VeSync, Apple, disclose.io) and news reporting (SecurityWeek, DroneDJ). It is not a substitute for a program's current terms or for legal review of bounty, release and safe-harbor language.
Put this guide to work
Frequently asked questions
What is the actual difference between a vulnerability disclosure policy (VDP) and a bug bounty program?
A VDP is a free, authorized channel for reporting security issues with safe-harbor protection for the reporter; a bug bounty program pays for valid findings. CISA's own Binding Operational Directive 20-01, which requires every US federal civilian executive-branch agency to publish a VDP, states this distinction directly: "a VDP is similar to, but distinct from, a 'bug bounty.' In bug bounty programs, organizations pay for valid and impactful findings," and the directive explicitly "does not require agencies to establish bug bounty programs" — a VDP is the mandatory baseline, a bounty is an optional, separately-governed layer an organization can add on top. For a device vendor, this maps directly: see [vulnerability disclosure policy template](/guides/vulnerability-disclosure-policy-template/) for the free, always-required baseline a bug bounty should sit on top of, not replace.
What does "scope" actually need to cover for a hardware or IoT bug bounty?
More than a software program does, because the asset itself can be physical. Real published programs make this explicit: Intel's Vulnerability Disclosure Program (run via Intigriti) states that valid submissions against firmware weaknesses requiring physical access or a UEFI shell are out of scope, while operating-system-level kernel exploits are generally in scope, and weaknesses in out-of-band management hardware such as BMC or CSME are generally in scope — three different scope decisions for three different layers of the same device. VeSync's IoT bug bounty program scopes in its mobile app plus the specific device categories it supports (outlets, switches, lighting, kitchen appliances, air purifiers, scales, trackers) rather than "all VeSync hardware," and Motorola Mobility's hardware-focused program (run via Bugcrowd) scopes to devices running specific Android versions and patch levels, explicitly excluding web-app-related findings. The common thread: a hardware/IoT scope has to name which physical access methods, firmware layers, companion apps, and device models are actually in play, not just a product line.
Does a bug bounty program's safe-harbor language actually protect a researcher?
Only to the extent the program's own written terms say so, and real programs vary. Arm's program (via Intigriti) states that ethical hacking consistent with its researcher guidelines is treated as "authorized" conduct, and that Arm will not pursue civil action for accidental, good-faith violations — but adds that it "is not able to grant authority or safe harbor to researchers for any security vulnerability testing of third-party software or systems," a limit that matters for any device running third-party components. Intel's safe-harbor waiver similarly "does not apply to your security research that involves the networks, systems, information, applications, devices, products, or services of another party (which is not Intel)." The practical rule, confirmed by multiple programs' own published terms: safe harbor covers only what the program operator actually owns or controls, and testing anything adjacent — a cloud backend run by a different company, a third-party chipset's own firmware — can fall outside even a program with strong, good-faith safe-harbor language.
What happens when scope and payout terms are unclear? Is there a real example?
Yes — DJI's 2017 "Threat Identification Reward Program" is a well-documented case, and a useful cautionary one precisely because DJI later improved. DJI launched the program with a one-page press release offering $100-$30,000 rewards but, per contemporaneous reporting (SecurityWeek, DroneLife), with no defined scope or rules published alongside it. Researcher Kevin Finisterre reported exposed SSL certificates and firmware AES keys and was told he qualified for the $30,000 maximum — but after his lawyers reviewed DJI's proposed release agreement and found it, in his account, "extremely risky" and possibly "crafted in bad faith to silence anyone that signed it," he declined the payment and published his own 18-page writeup instead, titled "Why I Walked Away From $30,000 of DJI Bounty Money." A separate researcher, Sean Melia, reported an unpatched Heartbleed flaw plus a SQL-injection/remote-code-execution issue with root access and was offered $500; he also declined, and DJI said he had broken a non-disclosure term by speaking to a reporter. Nine years later, in March 2026, DJI promised a $30,000 reward to software engineer Sammy Azdoufal for a critical flaw in its Romo robot vacuum's cloud backend (reported by DroneXL, DroneDJ and AndroidHeadlines, among others) with no reported scope or payment dispute — evidence that well-defined scope and release terms, not the size of the reward, are what actually determine whether a bounty program works for both sides.
Sources
- CISA Binding Operational Directive 20-01, Develop and Publish a Vulnerability Disclosure Policy (2 September 2020)
- Intel Vulnerability Disclosure Program rules (via Intigriti) — scope, firmware/physical-access and BMC/CSME terms
- Arm Bug Bounty Program rules (via Intigriti) — authorized-conduct and third-party safe-harbor limits
- VeSync Security Center — IoT Bug Bounty Program (scope: app + supported device categories; successor to VeSync's former standalone bounty page)
- SecurityWeek: Drone Maker DJI, Researcher Quarrel Over Bug Bounty Program (2017)
- DroneDJ: DJI fixes ROMO security bug that exposed thousands of homes (10 March 2026)
- Regulation (EU) 2024/2847 (Cyber Resilience Act), Official Journal text: Annex I Part II, Article 13(17), Annex II, recital (76), Article 71(2)
- ETSI EN 303 645 V3.1.3 (2024-09), Cyber Security for Consumer Internet of Things: Baseline Requirements (provisions 5.2-1, 5.2-2)
- UK SI 2023/1007, PSTI (Security Requirements for Relevant Connectable Products) Regulations 2023, Schedule 1 paragraph 2
- ISO/IEC 29147:2018 Vulnerability disclosure, catalogue entry and preview (contents: 9.3.5 Scope, 9.3.7 Recognition, 9.4.2 Legal considerations)
- US Department of Justice, Justice Manual 9-48.000, Computer Fraud and Abuse Act (good-faith security research)
- eCFR, 37 CFR 201.40(b)(18), DMCA exemption for good-faith security research
- disclose.io dioterms, safe-harbor policy language (CC0-1.0)
- Apple Security Research Device Program
- Apple Security Bounty categories and maximum rewards
- Tesla bug bounty on Bugcrowd (registration of researcher-owned hardware)
About this guide
SoftWSP Firmware Security Desk
Covers the PSIRT & disclosure cluster: 15 guides.

In this cluster
Questions about this page?
