Posted by Tim Willis, Project Zero
At Project Zero, we spend a lot of time discussing and evaluating vulnerability disclosure policies and their consequences for users, vendors, fellow security researchers, and software security norms of the the larger industry. We're very happy with how well our disclosure policy has worked over the past five years. We've seen some big improvements to how quickly vendors patch serious vulnerabilities, and now 97.7% of our vulnerability reports are fixed within our 90 day disclosure policy.
In saying that, it's a complex and often controversial topic that is frequently discussed both inside and outside of the team. We often receive feedback from vendors that Project Zero works closely with regarding our current policies: sometimes it's things they want us to change, other times it's how our work has positively impacted their work and users. Conversations like these have helped develop our policies over the years. For example, we introduced our 14-day grace-period in 2015 after helpful discussions with various vendors.
We recently reviewed our policies and the goals we hope to accomplish with our disclosure policy. As a result of that review, we have decided to make some changes to our vulnerability disclosure policy in 2020. We will start by describing the changes to the policy, and then discuss the rationale behind these changes.
Summary of changes for 2020
For vulnerabilities reported starting January 1, 2020, we are changing our Disclosure Policy: Full 90 days by default, regardless of when the bug is fixed.
Fix a bug in 20 days? We will release all details on Day 90.
Fix a bug in 90 days? We will release all details on Day 90.
If there is mutual agreement between the vendor and Project Zero, bug reports can be opened to the public before 90 days elapse. For example, a vendor wants to synchronize the opening of our tracker report with their release notes to minimize user confusion and questions.
We will try this policy for 12 months, and then consider whether to change it long term.
The current list of changes for 2020:
* The grace period is an additional 14 days that a vendor can request if they do not expect that a reported vulnerability will be fixed within 90 days, but do expect it to be fixed within 104 days. If a grace period is requested, and a bug is fixed between 90 and 104 days after it was reported, bug details will be released on the day it is fixed. Grace periods will not be granted for vulnerabilities that are expected to take longer than 104 days to fix. Note that the seven day deadline for vulnerabilities that are being actively exploited "in the wild" will remain unchanged.
We're constantly considering whether our policies are in the interest of user security, and we believe this change is a further step in the right direction. We also think it's simple, consistent and fair.
Rationale on changes for 2020
For the last five years, the team has used its vulnerability disclosure policy to focus on one primary goal: Faster patch development.
We want to make attacks using zero-day exploits more costly. We do this through the lens of offensive vulnerability research and evidence of how real attackers behave. This involves discovering and reporting a large number of security vulnerabilities, and through our experience with this work, we realised that faster patch development and patch deployment were very important and areas for industry improvement.
If patches take a long time to develop and deploy, then we quickly fall behind the curve: more bugs are introduced than vendors can fix and a herculean effort is required to get things back on track.
We also regularly uncover cases of "bug collisions" with our research. This is where the vulnerability we discovered was previously found and exploited by a real attacker. Knowing that the vulnerabilities that we find are often already being secretly exploited to harm users creates a sense of urgency, and so we ask vendors to fix issues as quickly as possible.
After five years of applying a 90-day disclosure deadline, we're proud of the results we've seen: vulnerabilities are being fixed faster than ever. For example, around the time Project Zero started in 2014, some issues were taking upwards of six months to fix. Fast forward to 2019, and 97.7% of our issues are fixed under deadline. That said, we know there is still room for improvement, both in industry-wide patch development speed and over the entire vulnerability management lifecycle.
Revisiting our underlying policy principles and goals
We recently spent some time articulating the underlying principles of our policies:
- Simple. Simplicity is important because we want to be easily understood. We also want to operate at scale, otherwise it's even harder to aggressively push all of industry to do better.
- Consistent. We want to be reliable and predictable. We want to apply our deadlines in a deterministic manner without fear or favour, demonstrating that we mean what we say. The bar for any exception/inconsistency needs to remain extremely high (note: we've only had two exceptions in the five year history of the team).
- Fair. We want to be equitable, balanced and impartial. We don't want to be in a position where different vendors (including Google!) get a form of preferential treatment. The same rules should apply to everyone.
We also realised as part of our discussions that there are two other policy goals that we wanted to include. Based on those discussions, here are our policy goals for 2020:
- Faster patch development (existing): We want vendors to develop patches quickly and have processes in place to get them into the hands of end users. We will continue to pursue this with urgency.
- Thorough patch development (new): Too many times, we've seen vendors patch reported vulnerabilities by "papering over the cracks" and not considering variants or addressing the root cause of a vulnerability. One concern here is that our policy goal of "faster patch development" may exacerbate this problem, making it far too easy for attackers to revive their exploits and carry on attacking users with little fuss.
- Improved patch adoption (new): End user security doesn't improve when a bug is found, and it doesn't improve when a bug is fixed. It improves once the end user is aware of the bug and typically patches their device. To this end, improving timely patch adoption is important to ensure that users are actually acquiring the benefit from the bug being fixed.
This new policy direction for 2020 gives clear incentives for vendors, especially those that have raised the following issues with us in the past.
- Since we founded Project Zero, some vendors hold the view that our disclosures prior to significant patch adoption are harmful. Though we disagree (since this information is already public and being used by attackers per our FAQ here), under this new policy, we expect that vendors with this view will be incentivised to patch faster, as faster patches will allow them "additional time" for patch adoption.
- The full 90 day window is available to perform root cause and variant analysis. We expect to see iterative and more thorough patching from vendors, removing opportunities that attackers currently have to make minor changes to their exploits and revive their zero-day exploits.
- We're also being explicit on improving patch adoption, since we're incentivising that vendors should be able to offer updates and encourage installation to a large population within 90 days.
We also really like that the new policy will improve the consistency of our disclosure process, while also remaining simple and fair. For example, some vendors considered our determination of when a vulnerability was fixed as unpredictable, especially when working with more than one researcher on the team at a given time. They saw it as a barrier to working with us on larger problems, so we're going to remove the barrier and see if things improve. We hope this experiment will encourage vendors to be transparent with us, to share more data, build trust and improve collaboration.
Disclosure policy is a complex topic with many trade-offs to be made. We don't expect this policy to please everyone, but we’re optimistic that it will improve on our current policy, encompasses a good balance of incentives and will be a positive step for user security. We plan to re-evaluate whether it is accomplishing our policy goals in late 2020.
No comments:
Post a Comment