To keep up with California's unrelenting wildfire threat, some insurers are now turning to AI to predict fire risk with unprecedented, structure-by-structure detail.

Why it matters: This will allow them to cover homes in areas that they would otherwise have passed over — but potentially at the cost of hiking rates for those who can least afford it.

The big picture: Spooked by a recent surge in destructive fires that shows no sign of cooling off, insurers have backed away from underwriting in the most flammable parts of the state. They say the risk is sky-high, and there's too much uncertainty about where fire will strike next and what it will consume.

• Premiums are ballooning in fire-prone zones, and insurers are dropping thousands of homes in high-risk areas, according to California's Department of Insurance (CDI).
• Traditionally, insurers rely on fire maps and proprietary predictions from catastrophe modeling companies. These bring in detailed information about a property's location — but often overlook characteristics of the property itself.

Now, some insurers are getting creative. They are trying to pack in as much data as possible: information from building permits, records and codes — and, increasingly, satellite photos and aerial imagery from drones and aircraft.

• Automatically analyzing super-detailed, top-down images helps insurers understand crucial, property-specific risk factors.
• Some important ones: how close a house is to vegetation, how flammable that brush is and what the house's roof is made of.

Driving the news: MetLife announced this week that it's working with a Bay Area startup, Zesty.ai, to use this type of data for property-level scoring.

• Zesty.ai predicts risk based on building information, aerial imagery, patterns gleaned from examining decades of wildfires and data from fire scientists.
• MetLife has been extra conservative in fire areas, VP Carol Anderson tells Axios, in part because it has relied on traditional maps to assess risk.
• When MetLife implements the new scoring system early next year, it hopes to bring in new customers and retain old ones, Anderson says.

Several startups have popped up to sell these new models to insurers, and some of the old guard are developing them, too.

• About half a dozen companies have already met with CDI about their new models, according to Ken Allen, deputy commissioner for rate regulation.
• And insurers, at an inflection point in how they approach fire risk, are increasingly interested, says Janet Ruiz of the Insurance Information Institute, an industry association.

But, but, but: Experts worry that property-level scoring can result in higher premiums for people living in high-risk areas, who are often on low or fixed incomes.

• Low-income homeowners may be unable to afford property updates that would drive risk factors down, like replacing roofing or clearing trees. And if rates go up, their property values could go down in response.
• New, more granular models could drive a bigger wedge between premiums, says Allen of the Department of Insurance, "so high risk pay more and lower risk pay less."
• "We are seeing insurance companies over-rely on technology, and the consumers are paying the price," says Emily Rogan of United Policyholders, a nonprofit that advocates for insurance customers.

What they're saying: "MetLife follows standard actuarial principles for ratemaking to ensure our rates are not excessive, inadequate or unfairly discriminatory," a spokesperson told Axios.

• Zesty.ai founder Attila Toth argues that it ultimately falls on regulators, not his company, to make sure that its risk models don't discriminate. But in a report last year, CDI said it "does not have the necessary authority to regulate how insurers underwrite residential property insurance."
• After this story was published, Toth shared the summary of an outside auditor's report that analyzed the Zesty.ai fire risk model and concluded that, if applied properly, it would not "result in rates that are unfairly discriminatory."

The bottom line: "Moving to risk-based rates is overall a positive thing to do, but it could have a negative effect on people currently in these high-risk areas," says Lloyd Dixon, a RAND researcher who last year published a detailed study of wildfire's impact on insurance in California.

Go deeper: Deciding whether to rebuild after fire

Editor's note: This story has been updated with details on the Zesty.ai actuarial report.

Tech giants, startups and academic labs are pumping out datasets and detectors in hopes of jump-starting the effort to create an automated system that can separate real videos, images and voice recordings from AI forgeries.

Why it matters: Algorithms that try to detect deepfakes lag behind the technology that creates them — a worrying imbalance given the technology's potential to stir chaos in an election or an IPO.

Driving the news: Dessa, the AI company behind the hyper-convincing fake Joe Rogan voice from earlier this summer, published a tool today for detecting deepfake audio — the kind that recently scammed a CEO out of $240,000.

• The new detector, which Axios is reporting first, is open source, so anybody can go through the code for free to understand and potentially improve it.
• But the company gets something out of it: The detector is built on a Dessa platform, which you have to download (without paying) to set it up.

The big picture: There's an all-hands scramble for better detectors, which generally require a lot of really good examples of deepfakes. Researchers use them to train algorithms that can tell if media was created by AI.

• Yesterday, SUNY Albany deepfake expert Siwei Lyu released a dataset filed with celebrity deepfakes.
• Earlier in the week, Google and Jigsaw — both owned by parent company Alphabet — released a large set of video deepfakes.
• And earlier this month, Facebook, Microsoft and the Partnership on AI teamed up with academic researchers to release more deepfake videos — and offer a prize to the team that uses them to make the best detector.

Unlike these datasets, which allow researchers to cook up their own detectors, Dessa is releasing a pre-baked system — which has advantages and risks.

• The company felt a responsibility to release an antidote after it made the realistic Rogan voice, says Ragavan Thurairatnam, Dessa's co-founder.
• "I think it's inevitable that malicious actors are going to move much faster than those who want to stop it," he tells Axios. The free detector is a "starting point" for people to push detection forward.

But, but, but: Thurairatnam acknowledged that an open-source detector could help a particularly determined troll create new audio fakes that fool it. That's because generative AI systems can be trained to trick a specific detector.

• He argues that the potential for creating better detectors outweighs the probability that someone will misuse Dessa's code.
• But Lyu of SUNY Albany says there's some reason to worry. "In principle, such code will help both but probably more for making better generators."

Go deeper: Researchers struggle with containing potentially harmful AI

About half of the world's militaries are now flying drones, according to a sweeping new study published this week, revealing the swift spread of a critical technology that used to be too expensive or sophisticated for most countries.

Why it matters: Increasingly robot-crowded skies mean that clashes involving drones — like the recent attack on a Saudi oil facility that the U.S. has blamed on Iran — are likely to become commonplace.

What's happening: From cheap, off-the-shelf quadcopters to enormous, missile-toting aircraft, flying drones are not only proliferating widely, but they're becoming steadily more integrated into militaries, according to the report from Dan Gettinger, co-founder of the Center for the Study of the Drone at Bard College.

• They are already changing the way countries project power over adversaries. Chinese drones are flying over the South and East China Seas, Russian drones are over Ukraine, and Iranian drones allegedly operate in Yemen and Syria.
• Despite the explosion of new players, the U.S., China and Israel still have the most sophisticated drone operations, Gettinger tells Axios. But new leaders, like Turkey and Russia, are emerging.
• China has less experience operating drones than its peers — but, as one of the largest suppliers of drones to other countries, it's likely learning vicariously, Gettinger says.

Between the lines: The study's focus on training and R&D programs in addition to drone arsenals — all gleaned from public information — reveals some militaries' deeper preparations for drone warfare.

• South Korea, for example, has about as many drones as you'd expect for a country its size, according to the report. But a look at its training programs shows something different: Its military intends to train thousands of small-drone operators.
• "South Korea recognizes that small drones are going to become ubiquitous on the battlefield of the future," Gettinger says.

What to watch: Big R&D efforts are underway in several countries to develop drone swarms — groupings of drones that can be flown by one remote operator, or even autonomously.

Preparing for an augmented reality future (Ina Fried - Axios)

Disinformation campaigns found in 70+ countries (Oxford Internet Institute)

Medical images left up online (Jack Gillum, Jeff Kao & Jeff Larson - ProPublica)

The gender gap in 6 charts (Gretchen Gavett & Matt Perry - HBR)

Revenge of the English major (David Deming - NYT)

A group of five "AI musicians" released an album yesterday.

The "performers" are programs that generate music based on patterns found in human tunes and reading material — "from Atwood to articles about teenage life and growth in metropolitan areas," according to a press release that itself may have been written by a robot.

What's happening: It's not very good.

But, but, but: The company that created the "musicians" got a $200,000 investment to make a new AI-generated album every month.

What's next: If synthetic media is the future, we're not quite there yet.