A faulty computerized risk-assessment program predicted the offenders could be released under the state's non-revocable parole law that took effect in January 2010.
The inspector general found that about 1,500 offenders were improperly left unsupervised, including 450 who "carry a high risk for violence." The offenders otherwise would have been released under traditional parole, which requires them to report in regularly and follow specific rules.
The new law was designed for less serious offenders. Under non-revocable parole, offenders don't report to parole agents and can't be sent back to prison unless they commit new crimes.
The Department of Corrections and Rehabilitation said it relies heavily on a computerized program because it must review the criminal histories of more than 160,000 inmates and more than 100,000 offenders on parole.
Auditors found the risk assessment was wrong for 23.5 percent of more than 10,000 offenders who were considered for non-revocable parole between January and July 2010. Some were scored too high and others too low, with the lower-scoring inmates eligible for unsupervised release.
Even after the computer program was altered, analysts determined it was wrong in 8 percent of cases.
"CDCR should not compromise public safety ... by understating offenders' risk of reoffending and releasing high-risk offenders to unsupervised parole," the report said.
The department disputed the inspector general's analysis and conclusion.
"Alleged 'errors' ... have in large part been corrected," Lee Seale, the department's deputy chief of staff, wrote in a rebuttal letter. "We reject the notion that the California Static Risk Assessment is flawed and dispute the evidence the OIG cites in support of this claim."
The version of the assessment reviewed by the inspector general has now been obsolete for over a year, Seale wrote, and the department will keep working to improve the program developed by the University of California, Irvine, Center of Evidence-based Corrections.
Seale said the program has saved money and cut prison overcrowding by keeping many parole violators from returning to prison—important developments given current events.
The report came as the state struggles to safely release less dangerous convicts and parolees to help combat a $10 billion budget deficit. Compounding the pressure, the U.S. Supreme Court ruled Monday that the state must reduce its prison population by about 33,000 inmates over the next two years to reduce crowding and improve care for mentally and physically ill inmates.
Gov. Jerry Brown signed a law this year shifting responsibility for tens of thousands of lower-level criminals to the jurisdiction of counties, though the shift can't take place until legislators or voters approve funding for local governments.
Non-revocable parole would end under the new law, but the inspector general's report left state Sen. Ted Lieu, D-Torrance, who requested the report, wondering if California can accurately predict which criminals are less dangerous.
"The report, to me, confirms my worst fears," he said. "They have dangerous parolees running around who should not be."
Lieu has repeatedly called for the department to end unsupervised releases, most recently two weeks ago when an ex-convict on non-revocable parole was charged with murdering two people in Southern California. He reiterated his plea based on the inspector general's report.
The report is the latest in a series of reviews questioning California's parole practices. The inspector general previously found the state failed to properly supervise paroled rapist Phillip Garrido, who pleaded guilty last month to kidnapping a young Jaycee Dugard and holding her captive in a backyard compound for 18 years.
Another report said the department should have sent a paroled San Diego County child molester back to prison before he raped and killed two teenage girls.
The inspector general's report cited problems with the non-revocable parole program that were originally reported by The Associated Press last year.
The AP found that because of the way the law was written, the department was releasing some unsupervised parolees who had been convicted of violent or threatening crimes.
The AP also reported there were early problems with the risk-assessment program. The department was forced to return 656 parolees to active supervision after learning that nearly 10 percent of parolees released without supervision had committed more crimes than officials previously believed.
Officials said then that the problem was fixed, though the inspector general's report says the improved assessment program wasn't immediately used at all locations statewide.
Wednesday's report found the high error rate under the computerized program resulted from multiple problems:
— It did not include prior parole violations or some crimes offenders committed as juveniles.
— Some offenders whose risk should have been assessed by human beings were instead left to the computerized assessment. Only about 4 percent of the more than 270,000 risk assessments of inmates and parolees were done by hand.
— The system relies in part on criminal histories compiled by the California Department of Justice. But about half the histories don't reveal what happened to criminal charges, for instance whether suspects were convicted or acquitted, because the information isn't submitted by the court system.
The system relies on 22 factors that are supposed to predict whether offenders are likely to commit new crimes. They include things like age, gender, gang affiliations, previous convictions, disciplinary problems in prison, and previous parole violations. It then uses a mathematical algorithm to assign a risk score.
Only those with low and moderate risk can be placed on non-revocable parole.
However, even a low risk predicts that 48 percent of those parolees are likely to be arrested for a felony, and 18 percent convicted of a felony, within three years. A moderate risk projects that 69 percent of parolees will be arrested and 31 percent convicted of a felony within three years of their release.
The program has been tested and proved scientifically valid, the department said in response. However, researchers found the program is weak in predicting future convictions and weak to moderate in predicting future arrests.
Lieu said state lawmakers were unlikely to have passed the law in 2009 had they known those odds.