Senator Lindsey Graham (R-S.C.) told an audience that, “We've killed 4,700,” with drone strikes. “Sometimes you hit innocent people, and I hate that, but we're at war, and we've taken out some very senior members of al-Qaida.”
Graham was laying out a standard defense of drone strikes: that you do what you have to to fight your enemies. The more startling revelation was the 4,700 part. No government official had given a number of people killed by drone strikes before.
What's unclear is where this 4,700 figure is coming from. Micah Zenko, who has studied the different estimates of drone strike casualties, had this to say on his blog:
My report, Reforming U.S. Drone Strike Policies, compiled the averages found within the ranges provided by New America Foundation, Long War Journal, and The Bureau of Investigative Journalism (TBIJ) and produced a number about 1,200 fewer.
It is notable that Graham’s estimate nearly matches the TBIJ’s highest estimated range for “total reported killed” in Pakistan, Yemen, and Somalia: 4,756. Either Graham is a big fan of TBIJ’s work, or perhaps he inadvertently revealed the U.S. government’s body count for nonbattlefield targeted killings.
So Zenko's best guess is that Graham spilled some government info. That's the best guess, though if true, Graham may have broken secrecy laws by leaking classified information. A spokesperson for Graham only muddled the issue, saying:
The senator "quoted the figure that has been publicly reported and disseminated on cable news."
So, does that mean that Graham didn't do anything wrong because the 4,700 figure was already public, or did he get the number from cable news. (It's not like stuff like that never happens: Mitch McConnell, the Republican leader in the Senate, was recently fooled by a satirical news report about Guantanamo detainees getting health benefits.)
Either way, it seems Graham probably said something he shouldn't have. What's ultimately more important, however, is whether the number is accurate. Given that it's on the higher end of previously cited estimates, I would guess that it is.