Sam Sabin
@samsabin.bsky.social
Axios cybersecurity reporter, taking it day by day ✨ | 📩: sam.sabin@axios dot com, signal: SamSabin.01 (no pitches!), she/her
Researchers find a flaw in Apple's Private Relay that exposes users' real IP addresses, which also affects users of Tor's OnionBrowser in iOS: www.404media.co/apples-priva...
Apple's ‘Private Relay’ Is Exposing Users’ Real IP Addresses
Researchers found a group of issues that mean Private Relay isn't actually protecting users' real IP addresses. 404 Media verified the issues do expose IP addresses.
404media.co
NEW: Cyberattacks on U.S. water supply now being reported in 12+ states and 100+ municipalities, though some may be unrelated to wave of suspected Iranian attempts. The nationwide threat follows years of ignored warnings and neglect to water systems. www.nytimes.com/2026/08/05/u...
Hacks on U.S. Water Supply Follow Years of Warnings and Neglect (Gift Article)
Federal and state officials are racing to address an assault on the nation’s water supply that they believe is the work of Iranian hackers.
nytimes.com
Demis Hassabis is leaving his role as CEO of Google DeepMind to be the unit's chairman and chief scientist for Alphabet (Ina Fried/Axios) Main Link | Techmeme Permalink
Anthropic says three of its models, including an internal research model, gained unauthorized access to real-world systems during internal cybersecurity testing (Sam Sabin/Axios) Main Link | Techmeme Permalink
an AI agent framework led an attack on Hugging Face's internal systems. Hugging Face then used AI (specifically GLM-5.2 👀) to detect the intrusion and reconstruct what happened. the AI vs. AI cybersecurity reality is starting to take shape. www.axios.com/2026/07/20/h...
Hugging Face says an AI agent carried out an end-to-end cyberattack
The attack is one of the first public cases of an AI agent conducting an operation from beginning to end.
axios.com
don't worry, i've been asking the White House about some of GPT-5.5-Cyber's benchmarks beating Mythos since yesterday. so far: silence www.axios.com/2026/06/23/w...
White House quiet on OpenAI's Mythos-like model
The seemingly straightforward model release raises questions about what actually triggered the Trump administration's concerns about Anthropic.
axios.com
OpenAI is updating GPT-5.5-Cyber so it's more capable and permissive for advanced security work. The move comes at Anthropic's Fable 5 & Mythos 5 remain in limbo with the Trump admin. more on @axios.com: www.axios.com/2026/06/22/o...
OpenAI rolls out more capable version of its cybersecurity model
Even as Anthropic's Fable remains in limbo, the race to get advanced AI models in the hands of defenders continues to heat up.
axios.com
China-based operatives used ChatGPT to try and influence online debates about AI data centers and Trump's tariffs, OpenAI said in a report today. www.axios.com/2026/06/10/o...
China-based operatives used ChatGPT to shape AI data centers and tariff debates
The campaigns show how pro-China actors are testing AI tools.
axios.com
Two months after deeming Mythos so powerful that no existing safeguards were sufficient, Anthropic is rolling out its first Mythos-class model — Fable 5 — to the public. Anthropic is also working on a program to vet access to its new Mythos 5 model. www.axios.com/2026/06/09/a...
Anthropic releases first Mythos-level model for general use
Fable 5 includes safeguards to help protect against hacking and bio-terrorism.
axios.com
Sen. Hassan has requested a briefing with CISA on the reported GitHub repository that a private contractor left exposed online containing internal agency passwords & files www.axios.com/2026/05/19/c...
Senator requests
Researchers uncovered an exposed GitHub repository that included internal credentials and files.
axios.com
small bit of news in my story today: an Anthropic spokesperson says the company recently held a briefing with a majority of states and the District of Columbia on supporting their critical cybersecurity efforts. www.axios.com/2026/05/19/a...
AI hacking era could hit hardest at state and local level
Small towns, schools and utilities lack the staff and basic cyber defenses needed to fend off such threats.
axios.com
Palo Alto Networks tells me the company found 75 bugs in its products in its first month of using Mythos + GPT-5.5-Cyber. Get ready for an influx of vulns from other vendors, too. My latest: www.axios.com/2026/05/13/p...
Major cybersecurity firm says new AI models uncovered 7x more flaws
Palo Alto Networks says it found 75 flaws in its products after starting to use advanced AI cybersecurity models.
axios.com
sorry but the way that so many stories are framing gpt-5.5-cyber as the first tiered access openai model for cyber defenders is wild to me — like you guys know they did this with 5.4 too, right? & announced this strategy nearly a month ago? from 4.14--> www.axios.com/2026/04/14/o...
OpenAI expands access to cyber AI as hacking risks grow
Facing similar risks as Anthropic, OpenAI is trying a different strategy.
axios.com
OpenAI is rolling out GPT-5.5-Cyber, a model for vetted cyber defenders who want to use the tool for do malware analysis, write proof-of-concepts for bugs and more. The release comes as safety tests suggests the model is nearly just as powerful as Mythos 👀 www.axios.com/2026/05/07/o...
OpenAI makes its rival to Anthropic's Mythos more widely available to cyber defenders
Recent security testing suggests that GPT-5.5. model is nearly as good at finding and exploiting software bugs as Anthropic's Mythos Preview.
axios.com
New on @axios.com: As part of meetings with tech/cyber cos. + tech trade groups last week, ONCD floated an AI security framework that was already in the works before Mythos. On the table: DoD red-teaming of AI deployments at the federa/state/local government levels. www.axios.com/2026/05/04/t...
Trump administration considering safety review for new AI models
In a post-Mythos world, the White House is re-evaluating its hard line against the AI security measures it once shrugged off.
axios.com
New: Tom Parker, a security services lead at IBM, has emerged as a potential contender to lead CISA after the most recent nominee withdrew himself from consideration for the role, five people familiar tell me: www.nextgov.com/people/2026/...
IBM security executive emerges as possible contender to lead CISA
Tom Parker doesn’t have prior government experience, characteristics the Trump administration may be seeking in its next pick to lead the cyber agency, a person familiar said.
nextgov.com
"While GPT‑5.5 didn’t reach Critical cybersecurity capability level, our evaluations and testing showed that its cybersecurity capabilities are a step up compared to GPT‑5.4." openai.com/index/introd...
Introducing GPT-5.5
Introducing GPT-5.5, our smartest model yet—faster, more capable, and built for complex tasks like coding, research, and data analysis across tools.
openai.com
Trump's science + tech adviser issued a memo today accusing China-based actors of "deliberate, industrial-scale campaigns" to distill and copy American AI models. www.axios.com/2026/04/23/u...
U.S. accuses China of "industrial-scale" campaigns to steal AI secrets
The accusation could complicate President Trump's upcoming visit to Beijing.
axios.com
exclusive on @axios.com: OpenAI has been briefing federal agencies, state governments and Five Eyes allies on GPT-5.4-Cyber — including a D.C. event Tuesday with ~50 government cyber pros, a source familiar tells me. www.axios.com/2026/04/22/o...
Exclusive: OpenAI briefs feds and Five Eyes on new cyber product
Officials have been clamoring to get their hands on the latest AI tools, which promise huge security gains.
axios.com
Scoop: CISA doesn't have access to Anthropic's Mythos model, two sources tell me. Interestingly, Anthropic briefed both CISA and NIST's Center for AI Standards and Innovation ahead of its public announcement, and NIST is testing Mythos. @axios.com www.axios.com/2026/04/21/c...
Scoop: Top U.S. cyber agency doesn't have access to Anthropic's powerful hacking model
CISA is not on the list of more than 40 organizations with access to Anthropic's Mythos Preview.
axios.com
scoop w/ @mccuri.bsky.social: The NSA has been using Anthropic's Mythos despite the Pentagon's instance that Anthropic is a "supply chain risk," according to two people. The NSA was one of the 40+ orgs that got initial access to the model weeks ago, one source added. www.axios.com/2026/04/19/n...
Scoop: NSA using Anthropic's Mythos despite blacklist
The government's cybersecurity needs are outweighing the Pentagon's feud with Anthropic.
axios.com
Sources say NSA is using Mythos Preview, and a source says it is also being used widely within the DoD, despite Anthropic's designation as a supply chain risk (Axios) Main Link | Techmeme Permalink
scoop w/ @mccuri.bsky.social: The NSA has been using Anthropic's Mythos despite the Pentagon's instance that Anthropic is a "supply chain risk," according to two people. The NSA was one of the 40+ orgs that got initial access to the model weeks ago, one source added. www.axios.com/2026/04/19/n...
Scoop: NSA using Anthropic's Mythos despite blacklist
The government's cybersecurity needs are outweighing the Pentagon's feud with Anthropic.
axios.com
hello from World’s Lift Off event in SF this morning where Sam Altman’s “proof of human” company is unveiling a new World ID protocol and a slew of new partnerships this AM (tinder, zoom, Docusign, Shopify & more). my story on the news here: www.axios.com/2026/04/17/w...
Sam Altman's "proof of human" company pushes into mainstream services
The company known for its iris-scanning identity verification technology is partnering with Tinder, Zoom and more.
axios.com
exclusive from me: Sen. Maggie Hassan (D-N.H.) is demanding info from four major AI voice cloning companies about their strategies to stop scammers from abusing their tools. www.axios.com/2026/04/16/h...
Exclusive: Sen. Hassan presses AI voice cloning companies on preventing scams
The letters are one of the first steps for lawmakers eyeing new legislation and regulations.
axios.com
NEW: We documented almost 100 cases of deepfake sexual abuse happening in schools around the world—with more than 600 children being victimized by their peers. Mostly boys in 25+ countries created explicit imagery of girls they often know from school over the last couple of years
The Deepfake Nudes Crisis in Schools Is Much Worse Than You Thought
An analysis by WIRED and Indicator found nearly 90 schools and 600 students around the world impacted by AI-generated deepfake nude images—and the problem shows no signs of going away.
wired.com
OpenAI rolled out its GPT-5.4-Cyber model on Tuesday and is giving access to anyone approved for the highest tier of its Trusted Access for Cyber program. more on @axios.com: www.axios.com/2026/04/14/o...
OpenAI expands access to cyber AI as hacking risks grow
Facing similar risks as Anthropic, OpenAI is trying a different strategy.
axios.com
pretty notable post-mortem from the Axios maintainer who says hackers tricked him via a bespoke Slack channel, fake microsoft team calls and initial communications that imitated the founder of a legit company
