Another AI model turned hacker story is doing the rounds, and this time the finger points at a fresh name. Gizmodo has published a report headlined “Uh-Oh. Which Company’s AI Model Is Reportedly a Hacker Now Too?”, signalling that yet another major AI developer’s system has reportedly been tied to hacking activity. The details available at the time of writing are thin, and Gizmodo’s question-style framing withholds the company’s identity from the headline itself. But the phrase “now too” is the tell: reports of frontier AI models being misused for cyber operations are no longer isolated embarrassments for a single lab — they are becoming a recurring feature of the industry’s news cycle.
Key takeaways
- Gizmodo reports that another company’s AI model is “reportedly a hacker now too”, implying a fresh case of a frontier model tied to hacking activity.
- The “now too” framing places the claim in a wider pattern: multiple AI developers have now faced reports of their models being misused in cyber operations.
- The company at the centre of the report is not named in the material available here, so specifics should be treated as claims attributed to Gizmodo’s reporting.
- Security researchers have long warned that agentic AI can compress the attack chain — reconnaissance, phishing, code generation — into cheaper, faster workflows.
- For developers and businesses, the practical response is model-level due diligence: understand each provider’s safeguards, monitoring and disclosure record.
- Expect growing pressure on AI labs to publish threat-intelligence reports, and on policymakers to formalise incident-disclosure expectations.
What Gizmodo’s AI Model Hacking Report Actually Says
According to Gizmodo, a major company’s AI model is “reportedly a hacker now too” — in other words, the system has allegedly been used to perform, or materially assist with, hacking activity. That single word “reportedly” matters. Coverage of AI misuse typically originates either from a developer’s own threat-intelligence disclosures or from external security researchers, and early reports are frequently refined as technical detail emerges. Gizmodo’s headline deliberately poses the company’s identity as a question, which is itself a comment on how routine these stories have become.
Because the available material does not name the firm, any specific attribution belongs to Gizmodo’s full report rather than to settled public record. What can be said with confidence is narrower but still significant: another frontier-model developer is now publicly associated with a hacking claim, joining a lengthening list. For readers tracking which systems have faced misuse allegations — and how their vendors document safety — our AI models database catalogues the capabilities and published safeguards of every major model.
Why “Now Too” Matters: AI Model Misuse Is Becoming a Pattern
The most telling part of Gizmodo’s headline is not the unnamed company but the word “too”. Over the past two years, several leading AI developers have publicly documented attempts by criminal groups and state-linked actors to misuse their models, publishing periodic threat reports and banning offending accounts. Each disclosure was once treated as a one-off scandal. Collectively, they now describe an industry-wide condition rather than a single lab’s failure.
That shift changes the questions worth asking. “Can a large language model be misused for hacking?” was settled some time ago — the answer, demonstrably, is yes. The live questions are operational: how quickly does a developer detect abuse, how transparently does it disclose what it found, and how effectively do its safeguards raise the cost of the next attempt? A report like Gizmodo’s, whichever company it concerns, is ultimately a data point about detection and disclosure as much as it is about the attacker who prompted it.
How an AI Model Becomes a Hacking Tool
None of the systems implicated in stories like this was designed to attack anyone. In practice, industry reporting describes a handful of recurring routes by which models end up doing offensive work. Attackers social-engineer the model itself, posing as penetration testers or academic researchers to talk it past its refusals. They abuse API access at scale, often through stolen or fraudulently created accounts. And, increasingly, they exploit agentic features — a model’s ability to plan multi-step tasks, call external tools and iterate on results with limited human input. These are the same capabilities that power legitimate AI coding agents, which is precisely what makes the dual-use problem so stubborn.
The broader concern, as security analysts frame it, is less that AI invents new attack techniques and more that it compresses old ones:
| Attack-chain stage | Traditional manual approach | AI-assisted approach (as described in industry reporting) |
|---|---|---|
| Reconnaissance | Hours of manual research on targets and infrastructure | Model summarises open-source information in minutes |
| Social engineering | Hand-written phishing lures, often with tell-tale errors | Fluent, personalised lures generated on demand |
| Malicious code | Requires skilled developers | Code drafted, debugged and adapted conversationally |
| Vulnerability analysis | Specialist expertise and slow triage | Model helps parse scan output and prioritise weaknesses |
| Scale | Limited by attacker headcount | Agentic loops can run many tasks in parallel |
Security teams are quick to note that the same properties cut both ways: defenders use identical models to triage alerts, review code and draft detections faster than they could before.
What AI Hacking Claims Mean for Developers and Businesses
For organisations building on AI, the sensible response to a headline like Gizmodo’s is not panic but due diligence. Model selection is now partly a security decision. Providers differ meaningfully in how they monitor for abuse, how resistant their systems are to jailbreaking, and how candidly they report incidents — and those differences rarely show up on a pricing page.
Teams deploying agentic systems should treat them like any other privileged software: log what the agent does, constrain the tools and credentials it can reach, and require human approval for sensitive actions such as sending messages, executing code against production systems or touching customer data. A vendor’s disclosure record is worth weighing alongside benchmark scores. A lab that publishes detailed accounts of misuse it has caught is, counterintuitively, often a safer bet than one that has never reported anything — silence can mean an absence of incidents, or an absence of detection.
Open Weights, Closed APIs and the Safeguard Gap
Stories like this one also sharpen a structural divide in the industry. Closed models served through an API give their developers a central point of oversight: suspicious usage patterns can be flagged, accounts banned and safeguards updated for every user at once. Self-hosted open-weight models offer no such lever — once the weights are downloaded, no vendor can revoke access or observe what the model is asked to do.
That is not an argument against open models, which bring real benefits in cost, privacy and control — trade-offs we quantify in our open vs closed AI cost study. But it does shift responsibility for misuse prevention from the vendor to the deployer. Organisations weighing that choice should factor governance overhead into the equation alongside infrastructure spend; our self-hosting vs API calculator is a useful starting point for the cost side of that decision.
What to Watch as the AI Hacking Story Develops
Several things will determine how significant Gizmodo’s report turns out to be. The first is confirmation: which company is involved, and whether the developer publishes its own account of what happened, as has become customary after high-profile misuse reports. The second is the nature of the activity — whether the model merely assisted a human operator or automated substantial parts of an intrusion with minimal supervision, a distinction that matters enormously for risk assessment.
The third is the policy response. Each new incident strengthens the case, already being made by security researchers and some lawmakers, for standardised disclosure of AI misuse — something closer to the norms that govern data-breach reporting. If “reportedly a hacker now too” headlines keep arriving at the current cadence, formal rules look increasingly likely to follow.
Frequently asked questions
Which company’s AI model does Gizmodo say is hacking? Gizmodo’s headline deliberately poses that as a question, and the material available here does not name the company outright. Until the full report and the developer’s own response are widely corroborated, the identity should be treated as a reported claim rather than established fact.
Has an AI model actually carried out a cyberattack on its own? Publicly documented cases across the industry have generally involved human attackers directing a model or automating parts of their workflow through it, rather than a system acting with genuine autonomy. The trend that worries researchers is the steadily rising degree of automation.
Are AI developers doing anything to stop model misuse? Major labs operate safety and threat-intelligence teams, publish periodic misuse reports, ban offending accounts and tune their models to refuse harmful requests. The recurring stream of incidents suggests those defences are imperfect, not absent.
Should businesses stop using AI models because of hacking reports? No — but they should treat model choice as a security decision. Reviewing a provider’s safeguards, logging API activity and limiting what autonomous agents can access are proportionate responses; abandoning the technology is not.
What does an AI model turned hacker actually mean in practice? It is shorthand for a model being used to perform tasks in the attack chain — reconnaissance, phishing copy, exploit code or automated intrusion steps — usually at the direction of a human operator rather than of its own accord.
The bottom line
Gizmodo’s report that another company’s AI model is “reportedly a hacker now too” is short on public detail but heavy on signal. Whichever lab it concerns, the pattern is now unmistakable: frontier AI models are attractive tools for attackers, their safeguards are being probed continuously, and disclosure has become part of every developer’s job description. For users and businesses, the moment calls for informed model selection and tighter controls on agentic systems, not retreat. For the industry, each new “now too” headline raises the bar for proving that detection and transparency are keeping pace with capability.
Sources: news.google.com. Reported August 06, 2026.
