One Article Review
| Source |  GoogleSec |
|---|---|
| Identifiant | 8400810 |
| Date de publication | 2023-10-26 08:49:41 (vue: 2023-10-26 13:08:08) |
| Titre | Increasing transparency in AI security |
| Texte | Mihai Maruseac, Sarah Meiklejohn, Mark Lodato, Google Open Source Security Team (GOSST)New AI innovations and applications are reaching consumers and businesses on an almost-daily basis. Building AI securely is a paramount concern, and we believe that Google\'s Secure AI Framework (SAIF) can help chart a path for creating AI applications that users can trust. Today, we\'re highlighting two new ways to make information about AI supply chain security universally discoverable and verifiable, so that AI can be created and used responsibly. The first principle of SAIF is to ensure that the AI ecosystem has strong security foundations. In particular, the software supply chains for components specific to AI development, such as machine learning models, need to be secured against threats including model tampering, data poisoning, and the production of harmful content. Even as machine learning and artificial intelligence continue to evolve rapidly, some solutions are now within reach of ML creators. We\'re building on our prior work with the Open Source Security Foundation to show how ML model creators can and should protect against ML supply chain attacks by using |
| Envoyé | Oui |
| Condensat | our even inspection model slsa the we about above actions activity actor add address addresses advanced after against all allow allowing allows almost already also answer any apishaving applicable applications apply applying arbitrary are around artifact artifacts artifacts:similarities artificial as:who assure attack attacks available back bad badge basis because been behaviors behind being believe benefit best between biases bound build builder building built businesses but bypassing can cannot cards cause chain chainbased chains challenges chart cloud code collaboration collection combat community compatibility components compromise compromised concern concerns confidence connect consensus consider consumers content continue control controls conventional costs could covered created creating creators critical current daily dangerous data datasets dependencies describing detect detectable detecting determine developer developersscaling development development: attack developmentan diagram did different difficult directly discourages discoverable does don during earlier easy ecosystem effective effort efforts either enabled encourage engineer enormous ensure establish every evolve exact example executable execution existing experimental exploits extensions fact fail features feedback feel file first following follows forward foundation foundations framework frameworks free from fully further furthermore future generally get github gitlab goes google gosst gpus/tpus guidelines guiding happens hard hardened harm harmful has have help highlighting how hub hubs hundreds ideas identifies identify identity implement incentivizing include includes includes:having including incorporating increasing industry information infrastructure injected injection innovations input inputs insights instruct insufficient integrate integrations integritysigning intelligence internet issue jumpstart just keys know known leak leaked learning lens let levels leveraged license lifecycle lifecycles like like:who lived llms slsa lodato long looking machine make malicious malware manage many mark maruseac meaning means meiklejohn metadata method mihai misinformation mlfor model models modelscode more most need needed new normally not now occurred offers often one open openid other output owners paramount part particular particularly path piece pipelines platform platforms please poisongpt poisoning possible potential practices pressing pretrained prevent prevents principle prior problem problems process produced producer producers production programs propose protect protecting protection provenance provide provides providing public publication published question questions quickly rapidly reach reaching recent reducing regular released replaced replacedmetadata repository require responsibly retrained reverse right rotate run safe saif same sarah save/load scale secrets secure secured securely security securityin seen services set severity should show shown sign signature signatures signed signing sigstore similar similarities similarity similarly simple simply since slsa smaller software solution solutions solve some source space special specific specification spot spread stages standard standards steal step strong such suggest supply support swapping tampered tampering target team tell testing them these those thought threat threaten threats threatstogether through throughout thus time times today todayby token tooling tools towards traced trained training transforms transparency treat tricked trust trusting trustworthy trying turned two types undertrained underway unfolding universally uploaded upon uptick use used useful users using usually vectors verifiable verification verify version vision vulnerabilities vulnerability wasn way ways welcome well went what whether which will with:digital within without work workload would years “build:” “dependency:” “trainer” |
| Tags | Malware Tool Vulnerability Threat Cloud |
| Stories | |
| Notes | ★★ |
| Move | 
|
L'article ne semble pas avoir été repris aprés sa publication.
L'article ne semble pas avoir été repris sur un précédent.