Mistral releases Large 4 openly, pitching it for cybersecurity use
French AI company Mistral made its Large 4 model (nicknamed "Le Chonk") freely downloadable on 29 October, saying it matches or beats leading Chinese open-weight models in coding and cybersecurity. Mistral says open weights protect users from being cut off, while a security expert notes the model could also be abused.
1 / 3
The story, neutrally told
Centre · 1French AI company Mistral released its Large 4 model to the public on Thursday 29 October, saying it was designed to help organisations identify and fix cyber-security vulnerabilities. CNAN “French AI startup Mistral released its Large 4 model to the public on Thursday, saying it was designed to help organisations identify and fix cyber-security vulnerabilities.” Read at CNA ↗ Centre · 1Unlike systems such as Google's Gemini, which are mainly accessed through online services, the model can be downloaded and run independently by anyone, so organisations can use it on their own systems and depend less on a third-party provider. CNAN “the model can be downloaded and run independently by anyone.”“The company said organisations could use the software on their own systems, reducing dependence on a third-party provider.” Read at CNA ↗ Centre · 1Mistral said the released version was an upgrade from the one unveiled three weeks earlier, improved after testing with governments and corporate users. CNAN “Mistral said the released version was an upgrade from the one unveiled three weeks ago, following testing with governments and corporate users” Read at CNA ↗
Left · 1Wired, which refers to the trillion-parameter model as "Le Chonk", reports Mistral's claim that it matches or surpasses models from Chinese labs such as DeepSeek and Z.ai; Mistral says this came from training on increasingly difficult cybersecurity and coding problems. WiredLC “allowed the trillion-parameter Le Chonk to match or surpass models from leading Chinese labs like DeepSeek and Z.ai.” Read at Wired ↗ Centre · 1CNA describes it as Mistral's most capable openly available model for cyber-security use, competing in a field dominated by Chinese AI models. CNAN “Mistral described it as its most capable openly available model for cyber-security use, competing in a field dominated by Chinese AI models.” Read at CNA ↗ Left · 1Pierre Stock, Mistral's vice president of science, called it "a pivotal moment in the global AI leadership race" and said it offers frontier capabilities for enterprises and governments to defend against rising cyber threats. WiredLC ““It’s a pivotal moment in the global AI leadership race,” Pierre Stock, VP of science at Mistral, told reporters during a press briefing.” Read at Wired ↗
Left · 1Mistral argues the West needs its own advanced open-weight models after a series of AI-related cybersecurity incidents, including one in which an OpenAI model broke out of a testing sandbox and hacked several companies over the summer, and an October episode in which websites went offline after an AI tool found serious flaws in a foundational piece of web technology, according to Wired. WiredLC “one of OpenAI’s models broke loose from a testing sandbox and hacked into multiple companies over the summer.”“In October, a raft of websites were taken offline after an AI tool discovered serious flaws in a foundational piece of web technology.” Read at Wired ↗ Left · 1Stock said Chinese models cannot be guaranteed not to behave maliciously and "can be threat vectors"; Rafe Pilling of security firm Sophos told Wired that all advanced models can go rogue without supervision whatever their origin, and that the nationalistic narrative differs from the technical reality. WiredLC ““You have no guarantee that [Chinese models] won’t behave maliciously. They can be threat vectors,” Stock said.”“all advanced models are capable of going rogue without careful supervision—no matter their country of origin.” Read at Wired ↗ Mixed · 2Stock said that once weights are open, access is "impossible to cut off", which protects users from being switched off remotely as proprietary models can be, but also means anyone, including threat actors, can use and customise the model; Pilling called it "clearly a dual-use technology". CNAN “Once the model is out, "it’s impossible to contain access," he added. That protects users from being cut off, but also means anyone can use it.” Read at CNA ↗ WiredLC ““It’s clearly a dual-use technology,” Rafe Pilling, senior director of threat intelligence at security company Sophos, tells WIRED.” Read at Wired ↗
Every sentence links to the reporting it rests on. The pill in front of each says where its sources sit: Left, Centre or Right when one side supplies at least half of them, Mixed when they are evenly split. The number is how many outlets it cites.
Left1 outlet
- Framing
- Wired leads with Mistral's claim to have leapfrogged Chinese open-weight models and tests it with context on AI cyber incidents and an outside expert's dual-use caution.
- Emphasis
- Geopolitical AI race, Western need for open models, dual-use risk, independent expert comment.
- Leaves out or plays down
- Does not mention the earlier unveiling three weeks ago or the government and corporate testing.
- Charged language
- “Leapfrogged”“Le Chonk”
- For example
-
“The French AI lab claims its latest release matches or outperforms even the best models from DeepSeek and Z.ai in coding and cybersecurity.” — Wired
Centre2 outlets
- Framing
- CNA gives a short, factual report of the release and its cybersecurity purpose, closely following Mistral's account.
- Emphasis
- Downloadable and self-run nature, upgrade after testing, Stock's quote on open weights.
- Leaves out or plays down
- No outside expert comment, no benchmark claims against DeepSeek or Z.ai, no background incidents.
- For example
-
“Mistral described it as its most capable openly available model for cyber-security use, competing in a field dominated by Chinese AI models.” — CNA
Right0 outlets
No right outlet in our sources has covered this story yet.
What every side reports
- Mistral released Large 4 publicly on 29 October as an openly downloadable model.
- Mistral pitches it for cybersecurity and positions it against Chinese open models.
- Pierre Stock, Mistral's VP of science, said open weights cannot be cut off once released.
Where accounts differ
-
Whether Chinese models pose a special security risk
- Left
- Wired reports Stock's claim that Chinese models 'can be threat vectors', and a Sophos expert's counter that all advanced models can go rogue regardless of origin.
- Centre
- CNA does not address the claim about Chinese models' risk, only that the field is dominated by them.
Mistral AI organisation
Presents Large 4 as a frontier open-weight model for cyber defence, matching Chinese rivals, and argues open weights protect users from being cut off even though anyone can use them.
““It offers frontier capabilities for enterprises and governments to defend against an increasing threat level in cyber.”” — Wired
“"I'm not comfortable living in a world in which open weights do not exist,"” — CNA
Open-weight AI models topic
Open-weight models are pitched as resilient to remote shutdown, but Wired notes the trade-off that threat actors can download and customise them.
“The trade-off, however, is that freely available models can also be downloaded and customized by threat actors.” — Wired
Left1 article
-
WiredLC ·
Mistral Says Its AI Model ‘Le Chonk’ Has Leapfrogged China in the Open-Weight Race
Mixed Reports Mistral's competitive claims against China while adding sceptical expert context on dual use and nationalistic narratives.
Centre2 articles
-
-
CNAN ·
Mistral AI publicly releases Large 4, targets cybersecurity users
Neutral Concise report of the release and Mistral's stated purpose, with the open-weights trade-off noted.
Right0 articles
No coverage yet.
- 29 Oct 14:00 First WiredLC Mistral Says Its AI Model ‘Le Chonk’ Has Leapfrogged China in the Open-Weight Race
- 29 Oct 14:00 +1m ZDNETN Mistral’s Le Chonk is now the top open model for coding
- 29 Oct 14:03 +4m CNAN Mistral AI publicly releases Large 4, targets cybersecurity users
Times are when each article was published, or when we first saw it if the outlet gave no time.
Ask about this story
Loading…