nikky

ChatGPT Now Makes Animated GIFs as Image Use Hits 4 Billion a Week

People now make 4 billion images a week with OpenAI’s tools, up from 3 billion in September. From this week, some of those images can move: ChatGPT will turn a typed idea or an uploaded photo into a looping animated GIF, on every ChatGPT plan.

The feature arrived on Wednesday with Images 2.6, OpenAI‘s new image model. OpenAI says it makes pictures 50% faster than the version before and keeps characters, products and brand details more consistent when you ask for changes. It is rolling out in ChatGPT on the web, desktop and phone apps, and in Codex, OpenAI’s coding tool, across all subscription tiers.

How the GIFs work

You ask for a looping GIF and describe what should move. OpenAI’s help page gives the example “Make a looping GIF of a paper boat bobbing on the water.” You can also upload a still photo and ask ChatGPT to animate it.

Once the GIF is made, you can change its speed, step through it frame by frame, save it or share it. Android Authority reports the speed control runs from half speed to double speed. To change the animation itself, you describe the edit in the chat. One limit: ChatGPT can’t edit an animated GIF you upload from elsewhere.

Before this, ChatGPT could not make a real animated GIF in one step, The Verge notes. OpenAI pitches it at marketers animating a mascot and people making reaction memes.

The number behind it

OpenAI says the 4 billion weekly images work out at more than half a billion a day across ChatGPT, Codex and the paid developer service. That is a rise of about a third since September. GIFs give people a new reason to keep generating.

For developers, OpenAI split the model in two. GPT-Image-2.6 Flare is the fast option and can produce a low-quality image in as little as six seconds. GPT-Image-2.6 Sunburst is the slower, higher-quality option. Both are paid by use.

What to watch

Turning a still photo into a moving one is fun with a pet or a doodle. It is less fun if someone does it to your photo without asking. A moving photo of a real person can be passed around as if it were genuine, and ChatGPT’s picture tools have already drawn scrutiny, from the “replica” image trend to fake election photos in the US midterms. OpenAI points to a system card, a document describing its safety tests, for the safeguards on Images 2.6.

Why it matters

A feature that once needed separate apps now sits inside a chatbot many people already use, on every plan. That will make animated AI images far more common in group chats and feeds, and makes it worth checking where a moving image came from before you trust it.

The post ChatGPT Now Makes Animated GIFs as Image Use Hits 4 Billion a Week appeared first on DailyAI.

OpenAI’s Agents Found 948 Ways to Take Over Linux, at $1,279 Each

OpenAI says a team of its AI agents found 948 separate flaws in Linux that let an ordinary user take full control of a computer, and built a working attack for every one. The run cost $1.21 million, or $1,279 per flaw. Bug-hunting contests have paid $20,000 to $70,000 for a single flaw of this kind.

The figures come from a research post OpenAI published on Wednesday. They describe one run in mid-October using an unreleased “cyber model”, a version of its AI built for security work. The company says it has already sent hundreds of reports and fixes to the people who maintain Linux, starting with the most serious.

Linux is the free operating system underneath Android phones, most cloud servers and the world’s supercomputers. Its core, the kernel, runs to more than 40 million lines of code. A flaw there can reach billions of devices.

What the agents actually found

Every one of the 948 flaws is what security people call a local privilege escalation. In plain terms: someone who already has a basic account on a machine, or has tricked their way onto it, can use the flaw to become “root”, the all-powerful administrator who can read every file and change anything.

That is serious, but it is not the same as a remote break-in. An attacker still needs a foothold first, such as a stolen login or a malicious app. The flaws turn a small break-in into a total one.

OpenAI ran the attacks against a standard desktop version of Linux with its normal protections switched on, not a weakened test machine. It used the same rules as Pwn2Own and TyphoonPWN, two contests where hackers are paid for working attacks.

How the numbers break down

The work ran as a production line of AI agents, each with one job.

First, “scout” agents picked out the parts of the kernel an ordinary user can reach. They chose 5,034 targets, about 2.8 million lines, or roughly 7% of the code. That step cost $3,426.

Next, “analyst” agents read those targets and listed 2,308 possible flaws within 12 hours, for $316,549.

Then “triage” agents tried to prove each one by actually triggering it in a sealed-off virtual machine. Within 30 hours they confirmed 1,724, about three in four, as The Hacker News also reported. OpenAI says only 3.1% of the original list came from the AI misreading the code.

Finally, exploit agents tried to turn the confirmed flaws into real attacks. They produced 1,169 working takeover attacks and 291 attacks that leak private memory. After removing duplicates that shared the same root cause, OpenAI counted 948 distinct takeover flaws and wrote a fix for each. Human researchers reviewed the results before anything went to the Linux maintainers.

A successful attack cost an average of $359 in computing to build. The biggest single bill was the attack-building stage, at $775,929.

The economics are the story

OpenAI applied the contests’ past prize money to its haul. At $20,000 to $70,000 per flaw, the 948 would be worth $18.9 million to $66.3 million on paper, about 16 to 55 times what the run cost. OpenAI calls its cost an upper bound and expects better tooling to bring it down.

A second test points the same way. Against kernelCTF, a Google reward programme that uses a stripped-down, hardened version of Linux, the system produced 21 takeover attacks for about $32,936. OpenAI estimates those would be worth around $1.5 million in rewards if accepted, roughly 45 times the cost. For comparison, it says kernelCTF has accepted 26 such attacks from everyone combined so far in 2026.

The models are also getting better quickly. In late July, an earlier OpenAI model, GPT-5.6 Sol, found 166 unique flaws of this kind for $513,690. The October run found 948, more than five times as many, even though the kernel had three more months of fixes in between.

OpenAI also says the agents used at least 10 attack techniques that outside kernel experts had not seen before. It has withheld the details for safety reasons.

The people who have to fix it

The flaws do not fix themselves. Linux is maintained largely by volunteers and company engineers who review every change by hand, and they were already under strain before this report.

At the Kernel Recipes conference this autumn, Greg Kroah-Hartman, one of the kernel’s most senior maintainers, said the project is now fixing 33 security issues a day, LWN reported. The number fixed per release has roughly quadrupled from a long-run rate of about 500, a jump he put down to AI tools. His message was “don’t panic”: the kernel’s processes are coping. But he also said the best models are still wrong about half the time, and described a batch of 79 AI-found kernel bugs in which only 15 were confirmed as real, and all 15 had already been fixed.

In August he went further for one corner of the kernel. In a post to the kernel mailing list, he said he would reject AI-written patches for the drivers/staging area, with an exception for tested security fixes, because “at least 1/3 of the results they generate are flat out wrong or harmful.”

OpenAI’s own post shows why that caution makes sense. In an early run, its system reported 200 successful takeover attacks. Human review found that 20 of them had reached root through a different, unknown flaw from the one they were assigned. OpenAI added an independent check afterwards.

The maintainers are moving fast regardless. OpenAI says that by the time its 10-day scan and review finished, 61 of the takeover flaws had already been fixed in the main Linux code, 43 had fixes waiting, and 299 were being discussed in public. OpenAI says many of those fixes came from other contributors working on the same bugs.

What it means for you

For most people the advice is unglamorous: install updates. Android phones, home routers, smart TVs and cloud services all run on Linux, and the fixes only help once they reach your device.

OpenAI’s own conclusion is blunter. It says that for now, a low-level account on a system that is behind on updates “should be considered functionally equivalent to root access for a well-resourced threat actor.” In other words, if someone capable gets any foothold on an unpatched machine, assume they can take all of it.

It is the latest in a run of AI security results. Last month AI helped uncover the PHP2Shell flaw, and Anthropic said Claude had proved 123 open-source libraries could not crash. OpenAI itself this week raised its bug bounty to $172,000 for the worst flaws in its own products.

Why it matters

The cost of finding a serious Linux flaw has dropped to about $1,300 of computing, well below what contests have paid for one. The same capability that OpenAI pointed at defence will, as OpenAI itself expects, reach attackers too, so the real test is whether volunteer maintainers and device makers can ship fixes as fast as AI can find the bugs.

The post OpenAI’s Agents Found 948 Ways to Take Over Linux, at $1,279 Each appeared first on DailyAI.

US to Spell Out the Laws AI Labs Must Follow Before Congress Acts

Within the next two weeks, the US government plans to publish a list of every existing law and regulation that applies to American AI labs. The aim is to head off 535 members of Congress writing new rules of their own, a day after Democrats won control of both chambers.

Emil Michael, the Pentagon’s chief technology officer, gave the timeline on Wednesday at a Wall Street Journal technology conference in Napa, California, Reuters reported. He said the government’s Super Intelligence Task Force would set out the rules “in the next week or two”, but gave no further detail.

“What we’re trying to do is create a framework that makes the American people have confidence that these companies are going to do the right thing,” Michael said. “Or else I think what you get is 535 legislators writing a piece of regulation that could kill the industry.” That figure is the full membership of the House and Senate.

Who is behind it

President Trump set up the task force in October, NBC News reported. It is led by Director of National Intelligence Jay Clayton and includes Federal Trade Commission chair Andrew Ferguson, with Michael as vice chair. It reports to Trump and his chief of staff, Susie Wiles. It followed a White House meeting with tech leaders that Trump said produced “tremendous self-regulation”, a deal that does not appear to be legally binding.

What the labs have agreed so far

Michael said AI companies have signed up to four safety principles, including outside auditors and an independent board committee that can take safety complaints, according to Benzinga. He also wants rules on how companies report and sort safety incidents.

Asked whether the task force can pause a model or fine a company, he pointed to powers that already exist. Export controls, he said, could cut off a company’s foreign sales if it failed to fix a jailbreak, a trick that gets a chatbot past its safety rules, within two hours.

The pressure from Congress

Democrats are promising tougher scrutiny. Texas Representative Greg Casar said AI company bosses “should be called to testify under oath”, The Hill reported, and Senator Ed Markey joked about printing subpoenas. Any new law would still need Trump’s signature or enough votes to override a veto. The AI industry itself is split, with OpenAI among those pushing for some federal limits.

Why it matters

A list of existing laws is not new regulation, and it will not say how those laws are enforced against AI. But it will be the first time the administration has said in one place what AI labs can be held to, and it sets the starting point for whatever the new Congress tries next.

The post US to Spell Out the Laws AI Labs Must Follow Before Congress Acts appeared first on DailyAI.

DailyAI Briefing: Anthropic’s $17bn quarter ends in a loss, plus 4 more

Welcome to the DailyAI Briefing, the day’s five biggest AI stories. Anthropic’s updated IPO filing shows a $17 billion quarter that still ended in an operating loss, because rushed computing power cost far more than planned. OpenAI says ChatGPT now handles a fifth as many searches as Google, Google is making schools switch off Search if they want AI off, China is speeding up a single AI law, and stolen AI keys are running up million-dollar bills.

Let’s dig in.

Anthropic’s $17 billion quarter ends in a loss

$567 million. That’s the operating loss Anthropic reported for the three months to 30 September, on revenue of about $17 billion, in the amended IPO filing it sent to the US Securities and Exchange Commission on Monday. In the June quarter it made an operating profit of $974 million. The filing blames “$5.0 billion of compute under short-term commitments, which currently costs significantly more” than computing power bought on multi-year deals, plus $2.0 billion of non-cash charges from its staff donation-matching programme.

Computing and infrastructure costs came to $9.9 billion, about 58% of revenue, up from 48% the quarter before. Quarterly revenue growth also slowed to around 47%, from 144% in the second quarter, CNBC noted. The filing lists about $895.4 billion of committed spending, almost all of it on compute, through to 2045. It also says Anthropic’s own forecasts suggest it “may imminently meet” its safety policy’s threshold for “dramatically accelerated progress attributable to AI”, though no current model has been judged to cross it. The plain reading: demand for Claude is outrunning the chips Anthropic signed up for, and buying the gap at short notice is expensive.

OpenAI says ChatGPT runs a fifth as many searches as Google

One for every five. That’s how many search-like queries ChatGPT now handles compared with Google, according to Adam Fry, who leads ChatGPT’s consumer product, speaking to Big Technology. He also said weekly users rose from 1 billion to 1.2 billion in two months. OpenAI didn’t say how it counted Google’s side, and every web lookup inside a chat counts as a separate search.

Outside estimates are much lower. seoClarity puts it at about 7.5%, and Ahrefs estimated around 12% in February, while finding that Google sends roughly 190 times more visitors to websites. For anyone who runs a site, the clicks matter more than the searches, and those are still small. Our story sets the figures side by side.

Google makes schools choose between AI and Search

Gemini and Google Search “will be entirely off for K-12 student accounts until school administrators explicitly turn them on”, Google’s education chief Chris Phillips wrote on Tuesday. Schools can switch Search back on without the Gemini apps, but Search still brings AI Overviews with it. “Search and AI are deeply integrated,” Phillips said, according to Education Week.

Making AI opt-in for children is a sensible default. The trouble is that Google already offers a Web filter that shows only text links without AI Overviews, so a link-only school setting is clearly possible. In our opinion piece we argue Google should offer exactly that, so a cautious school doesn’t have to give up search to say no to AI.

China speeds up a single AI law

More than 700 million people in China now use generative AI, over half the population, according to official data released in September. On Tuesday, Wang Ruomeng of the National Development and Reform Commission, China’s state economic planner, said Beijing will “accelerate” one comprehensive AI law, at a press conference in Beijing. According to the Beijing Daily, it would come with rules for data, computing power and algorithms, incident reporting for AI firms, monitoring of AI’s effect on jobs, and protections against deepfakes and emotional dependence on chatbots.

There’s no draft text, deadline or penalties yet, and a draft AI law has sat on the State Council’s work plan since 2023, Caixin notes. Still, the job-monitoring and incident-reporting parts could produce public numbers on AI’s harms that few governments publish. We covered the details this morning.

Stolen AI keys run up million-dollar bills

More than $100,000 a day, and $1 million in total at some firms before anyone noticed. That’s what criminals have cost companies by stealing their AI access, according to a report from Google Threat Intelligence Group and Mandiant. Attackers take leaked cloud logins and API keys, switch on AI services the victim never used, and resell the access. One reseller, Jet Night, made more than 500,000 AI calls on a single victim’s keys in under eight hours and sells an unlimited plan for about $60 a month.

Request volumes can jump past 1,000 a second, BleepingComputer reported, and the AI provider simply bills it as normal use. Google’s advice is basic: rotate keys, scan code for leaked secrets, and set spending caps and alerts. If an AI deal looks far cheaper than the official price, someone else may be paying for it. Our story explains how the scheme works.

Signing off

Today kept coming back to the bill. Anthropic is paying a premium for compute it didn’t plan for, companies are paying for AI they never used, and schools may pay for opting out of AI with their search engine. Meanwhile, ChatGPT is counting its searches in a way that makes it look closer to Google than outside data suggests.

See you tomorrow.

The post DailyAI Briefing: Anthropic’s $17bn quarter ends in a loss, plus 4 more appeared first on DailyAI.

Stolen AI Keys Ran Up $1 Million in Bills, Google Says

Criminals who steal companies’ AI access are running up bills of more than $100,000 a day, and at some firms the total reached $1 million before anyone shut them out. That is the headline figure from a new Google report on a crime called LLMjacking.

The report, published on 2 November by Google Threat Intelligence Group and its incident-response arm Mandiant, covers attacks Mandiant handled in 2026. Victims were in finance, retail, technology and transport across North America, South America and Asia Pacific.

What LLMjacking is

A large language model (LLM) is the kind of AI behind chatbots. Companies pay for it by usage, through API keys, which work like passwords that let software talk to the model. LLMjacking means stealing that access and using it, or selling it, while the victim pays the bill.

Google compares it with cryptojacking, where hackers quietly use someone else’s computers to mine cryptocurrency. The difference is speed. Cryptojacking drains resources slowly. LLMjacking “can generate large spikes in fraudulent cloud charges in a matter of hours”, the report says.

How the keys get stolen

The most common way in was stolen or exposed cloud logins, including keys left in public GitHub code. One group tested stolen keys with tiny requests asking for a single word of output, so the checks would not set off alarms. Another built a fake sign-up page that tricked staff into pasting a live access token, which let the attackers skip passwords and two-factor checks.

Once inside, attackers switched on AI services the company had never used and created extra keys. Request volumes could then jump past 1,000 a second, BleepingComputer reported, and the AI provider simply logs it as normal paid use.

The $60 ‘unlimited’ plan

Much of the stolen access is resold through what Google calls “LLM Mixers”, services that pool stolen keys and sell cheap AI access. One, called Jet Night, made more than 500,000 AI calls in under eight hours on one victim’s keys and passed on nearly $1 million in costs. It sells an unlimited plan for about $60 a month. Google says a comparable legitimate service would usually cost $100 to $250, and believes Jet Night only makes money because it runs on stolen keys.

Google’s advice for companies is basic: rotate keys automatically, scan code for leaked secrets, give accounts only the permissions they need, and set spending caps and alerts on AI services. It also flags a flaw, CVE-2026-40217, in LiteLLM, a popular tool for routing AI traffic, which one group used to turn ordinary keys into admin access.

Why it matters

AI bills are now big enough to be worth stealing. For readers, the lesson is simple: an AI deal that looks far cheaper than the official price may be paid for by someone else’s stolen account. Companies that pay for AI, and reward people for finding its security bugs, now need spending limits as much as passwords.

The post Stolen AI Keys Ran Up $1 Million in Bills, Google Says appeared first on DailyAI.

Google Shouldn’t Make Schools Choose Between AI and Search

Google is giving schools an off switch for its AI tools. The catch is that the same switch turns off Google Search for students. In our view, that is the wrong trade, and Google already has the tool to avoid it.

On Tuesday, Google’s education chief Chris Phillips wrote on Google’s blog that Gemini, Gemini Notebook, Gemini in Classroom and Google Search “will be entirely off for K-12 student accounts until school administrators explicitly turn them on”. K-12 means primary and secondary school, roughly ages 5 to 18. Google says it will share details with schools in the coming months before the change rolls out.

This is an opinion piece. The facts below are sourced. The argument is ours.

What Google actually changed

Until now, Gemini was switched on by default in Google Classroom, so children could use it unless their district turned it off, NBC News reported. That drew pushback. In early October, 13 parent advocacy groups wrote an open letter asking districts to switch off Gemini and YouTube.

The new default is the reverse: everything starts off. After criticism on Tuesday, Google clarified that schools can turn on Search without the Gemini apps. But Search itself still comes with AI features such as AI Overviews, the AI-written summaries at the top of results. A plain, link-only Search is not on the menu. “If a school is not comfortable with that and simply wants AI off, then they have the choice to not have Search on at all,” Phillips said at a 2 November meeting, according to Education Week. “Search and AI are deeply integrated.”

Google’s Classroom product lead, Brittany Siegal Mennuti, told the same meeting that settings can differ by grade. A district could give older students full Search with AI summaries and limit younger ones to Guided Learning, a mode that walks students through a problem step by step instead of handing over the answer. Any group with AI switched off loses Search.

Why the bundle is the problem

Giving schools control is the right call. Making AI opt-in for children is a sensible default, and Guided Learning may well be useful. The problem is the price of saying no.

Google’s own help page says something that cuts against “deeply integrated”. It offers a Web filter that “displays only text-based links without features like AI Overviews”. So Google can already serve Search without the AI layer. It does so for any adult who clicks one tab. It is choosing not to offer that version as a school setting.

That choice matters because of Google’s reach in schools. Nearly two-thirds of assignments use a Google product, according to a 2023 Schoolytics report cited by NBC News. A district that wants no AI for its youngest pupils now has to pick between AI it didn’t ask for and no Google Search on school accounts. American Federation of Teachers president Randi Weingarten called it putting business interests ahead of children’s education. Fairplay’s Josh Golin said Google is trying “to strong-arm schools into using its AI products”.

Google says the change gives “schools more choice, not less”. That is true compared with August. It is less true compared with what Google could ship tomorrow. Search is still how most people find things online, even as ChatGPT claims a growing share. Students who never learn to read a page of plain results miss a basic skill.

What Google should do

The fix is small. Add a third school setting: Search with the Web filter on and AI features off. Google has built the product. It would let cautious districts keep Search while they decide on AI, and it would make the opt-in a real choice. Google told Engadget the products can be enabled individually and that it is “not an all-or-nothing deal”. A link-only Search setting would make that true for Search as well.

Why it matters

Defaults decide how millions of children first meet AI. If saying no to AI costs a school its search engine, few will say no, and that makes the opt-in look more like a choice than it is.

The post Google Shouldn’t Make Schools Choose Between AI and Search appeared first on DailyAI.