Secret technology

Dual_EC_DRBG: the NSA-backed random number generator and RSA’s $10 million deal

A random number generator promoted by the NSA had a built-in trapdoor, and RSA was reportedly paid $10 million to make it the default. NIST pulled it in 2014. Did the NSA hold the key?

RedactedReal facts underneath, and the theory fits them, but pieces are still blacked out.

Published Oct. 7, 2026Last checked Oct. 25, 2026

What’s claimed

Every encrypted connection depends on random numbers nobody can guess. The claim is that the NSA steered one particular generator, Dual_EC_DRBG, into an official US standard while holding a secret that lets its owner predict the supposedly random output, and then paid the security company RSA $10 million to make it the default in a widely used toolkit. If true, anyone relying on that default was handing the agency a skeleton key.

What’s on the record

Dual_EC_DRBG was one of four generators in NIST Special Publication 800-90, published in June 2006. It is built on two fixed points on an elliptic curve, called P and Q, and is far slower than the alternatives. In August 2007 two Microsoft researchers, Dan Shumow and Niels Ferguson, showed that anyone who knows the mathematical relationship between P and Q can, after seeing a small amount of output, predict what comes next. The cryptographer Matthew Green of Johns Hopkins has noted that nobody outside knew where the recommended points came from.

In September 2013 the New York Times, drawing on Snowden documents, cited Dual EC as an example of the NSA’s Bullrun programme, and NIST immediately recommended against using it. On 21 April 2014 NIST removed it from its draft revision of the standard, citing low public confidence and urging users to switch to the remaining three generators.

On 20 December 2013 Reuters reported that the NSA had paid RSA $10 million to make Dual EC the default generator in its BSAFE toolkit. RSA said it had chosen the default in 2004 and stated: "We categorically deny this allegation." Ed Felten of Princeton called it an artfully worded quasi-denial: it disputed a secret deal to weaken RSA’s products, not that the contract existed, and he asked why RSA kept the default after the 2007 warning.

In 2015 Michael Wertheimer of the NSA wrote in the Notices of the American Mathematical Society that, with hindsight, the agency should have stopped supporting the algorithm as soon as the trapdoor potential was shown, and called its failure to do so "regrettable." In December 2015 Juniper Networks disclosed that some versions of its ScreenOS firewall software used Dual EC, and that an unknown party had changed the Q point and some test values.

What doesn’t add up

No public document shows that the NSA holds the secret number behind Q, or that it ever used one. Felten thought it likely RSA did not know of the weakness when it signed; he also noted that NIST never named a default and that RSA appears to have been the only major supplier to choose this one. Researchers collecting the evidence at projectbullrun.org point to patent filings by the company Certicom suggesting the back-door idea was understood by 2005, before the standard was final, and to a 2007 change that made the trapdoor usable in more settings. That is evidence of what was possible and known, not proof of what was done.

Still open

No public document shows where the points P and Q came from, whether anyone holds the secret number behind Q, or whether it was ever used. Nobody has said who changed the Q point inside Juniper’s ScreenOS, disclosed in December 2015; it is still an unknown party. And Ed Felten’s question stands: why did RSA keep the default after the 2007 warning? A document tracing the points to their maker would move this file.

Where it stands (for now)

Redacted. This one sits at the strong end: the trapdoor is mathematically real, NIST withdrew the standard, an NSA official called the agency’s support regrettable, and RSA never squarely denied the reported payment. That the NSA holds the key, and used it, has not been shown on paper so far.

Sources

  1. Document NIST Removes Cryptography Algorithm from Random Number Generator Recommendations · National Institute of Standards and Technology · April 20, 2014
  2. Source The Many Flaws of Dual_EC_DRBG · A Few Thoughts on Cryptographic Engineering (Matthew Green) · Sept. 17, 2013
  3. Source RSA doesn’t quite deny undermining customers’ crypto · Center for Information Technology Policy, Princeton University (Ed Felten) · Dec. 22, 2013
  4. Source Dual EC DRBG · projectbullrun.org
  5. Source Dual_EC_DRBG · Wikipedia

More: the case files, every source behind every story.

Talk about it

Start a thread on this story Know more? Send a tip